#CLI Changelog

2026-10-30

#0.119.0

#New Features

• tessl loop - The new name of the tessl schedule command group. Every subcommand works under both spellings, and help, hints and output say loop
• Environment bindings - tessl environment bind <name> and tessl environment unbind <name> bind the linked project, or --project <name>, to an organization environment, or remove the binding. Both take --org and --json. tessl environment create names the bind command when it creates an environment no project can use yet

#Improvements

• Requested by - tessl loop runs <id> shows who asked for each run beside who it runs as: a person, an API key, a schedule, another run, a GitHub, Linear or Slack user, or a finding
• Project automation as creator - tessl loop list shows Project automation <project> as the creator of a Loop defined in tessl.json, and tessl loop runs <id> shows it for the runs such a Loop starts
• Integrations in help - tessl integrations list and tessl integrations show <type> appear in tessl --help
• Cloud launch flags in help - tessl launch skill --help lists --environment and --project
• Loop terminology - tessl loop list heads its trigger column Triggers, tessl loop pause and tessl loop delete say that runs already started continue, and every loop command says loop in its messages
• Lens limit per organization - tessl code review runs up to 12 lenses from a repository review profile or --skill, up from 8, and an organization can be set to a different limit. A review that applies more lenses than the limit is refused with a message naming it. The built-in standard profile is unaffected

#Bug Fixes

• tessl environment update --env-file keeps an organization environment's secret references. Before, it removed every reference. A name the file sets becomes a plain value and its reference is dropped, and the command names the converted variables
• tessl loop view <id> shows the environment a Loop names when the project is no longer bound to it, with the reason, instead of Default (no named environment)
• tessl loop runs <id> shows API key deleted for a run created by a revoked or deleted key, and shows the key for an older run that recorded both a key and its owner
• tessl plugin unpublish no longer warns that unpublishing works only within 2 days. The registry refuses only when the plugin is public, the version is more than 48 hours old, and the plugin has 500 or more downloads

2026-10-26

#0.118.1

#Bug Fixes

• tessl --version and tessl -v print the version without first checking for or installing an update
• Installing a plugin on Windows no longer fails with EPERM when another process, such as an antivirus scanner, briefly holds one of its files open. The CLI retries the move for about 2 seconds

2026-10-23

#0.118.0

#Improvements

• New GitHub App names - tessl github token, tessl github setup-auth and the bundled tessl/harness-engineering workflow guidance call the read-write GitHub App Tessl App, its new name on GitHub, formerly Tessl Helper. The guidance gives its bot login as tessl-app[bot]
• Tessl App in integrations - tessl integrations list and show label the read-write GitHub App connection GitHub (Tessl App), formerly GitHub (Tessl Helper)
• Waiting runs in schedule trigger - tessl schedule trigger says when the run it started is waiting for its debounce window or for another run on the same concurrency key to finish. --json output and the schedule_trigger MCP tool carry waiting as debounce or admission

#Bug Fixes

• The CLI now removes emSign ECC Root CA - G3 from its built-in trusted root certificates, so its own HTTPS connections no longer accept certificates that chain to it. Mozilla removed this root after the private keys of eight eMudhra intermediate certificates issued under it were reported compromised.

2026-10-21

#0.117.0

#New Features

• tessl project delete - Delete a project by its ID, its workspace/project reference or its name. Run history is kept. It asks for confirmation in an interactive shell, and --yes skips it
• Follow a launch run live - tessl launch logs <id> --watch prints each new event as one JSON line until the run finishes, and keeps going through transient errors and retried attempts
• Paging for launch runs - tessl launch list --after <cursor> fetches the next page, and --json output carries the next page's cursor as nextCursor
• Issue links for launch runs - tessl launch list --link-type accepts github_issue
• Skill version in launch view - tessl launch view shows the registry plugin version a run installed

#Improvements

• Eval failure reasons - tessl eval run and tessl eval view print why each failed solution failed, grouped by reason
• Clearer eval help - tessl eval --help lists its subcommands and the commands that show results, and --variant-json help describes the variant format
• Uninstallable dependencies in doctor - tessl doctor --automations reports a file: dependency that is an absolute path, resolves outside the repository, or is not committed, because a cloud run cannot install it
• Free Skill Review - tessl review run no longer blocks an organization that is out of credits, because Skill Review is now free
• Launch view branch - tessl launch view shows repository default for a run that names no base branch

#Bug Fixes

• tessl verify during an uncommitted merge measures only the branch's own changes. Before, merging main into a branch counted everything main brought in
• tessl environment list and the tessl environment delete picker show each organization environment once. Before, an organization environment bound to the workspace's projects appeared twice, and picking the second copy failed to delete it
• Large --json output piped to another program is no longer cut short when the CLI exits
• tessl scenario generate on a plugin sends its existing evals, so generated scenarios no longer reuse an existing scenario's directory name
• tessl scenario download and the tessl eval run guard point at the plugin root when the output is inside its evals/ directory

2026-10-16

#0.116.0

#New Features

• Plugin and Markdown lenses in code review - With no repository profile and no --skill, tessl code review sends files under a directory containing .tessl-plugin/ to a plugin content lens, other Markdown files to a Markdown lens, and everything else to the four code lenses. The default lenses are pinned to tessl/code-review@0.5.0
• Project filter for schedules - tessl schedule list --project-id <uuid> lists only the schedules of that project
• Browser links for projects and environments - tessl project create, link and repair print a Web URL for the project, and tessl environment create and view print one for the environment. tessl project repair --json and tessl environment view --json include webUrl
• Browser links for schedules and launch runs - tessl schedule trigger and tessl launch view print a browser link, tessl schedule list and tessl launch list print the link pattern, and the --json output of all four adds webUrl
• Organization environments - tessl environment create creates an environment for your organization and binds it to the project from --project, or to the linked project. --org replaces --workspace. The other environment commands look a name up in the organization first, then in the workspace
• Project on environment runs - --environment on tessl schedule create and tessl task create sends the run's project, from --project or the linked directory. In a directory linked to no project, pass --project

#Improvements

• Schedule validation covers actions - Offline, tessl schedule validate reports a schedule whose action names an action the file does not declare, and problems in the actions block, and exits non-zero. In --json, an actions problem carries "block": "actions"
• Trigger errors in schedule view - tessl schedule view prints the error the API reports for each trigger, with its time
• Clearer schedule view - tessl schedule view shows Default (no named environment) when a schedule names no environment, and labels a full commit SHA Commit: instead of Branch:
• Faster plugin hooks - The commands that plugin hooks run on every agent tool call no longer start analytics, send trace spans or claim cache ownership at startup, so each hook starts faster
• Bundled code review plugin - tessl agent bundles tessl/code-review 0.5.0, which adds the plugin content and Markdown lenses
• tessl environment list shows the projects bound to each environment. Its --json output now has environments and workspaceEnvironments

#Bug Fixes

• A plugin hook script that exits 2 with nothing on stdout now blocks the agent's action, using the script's stderr as the reason. Before, the agent received an allow
• A schedule's local skill path (./, ../ or file:) is read from the directory of its project's tessl.json for every schedule, as tessl schedule create --skill help now says. Before, a schedule created through the CLI or the web app read it from the repository root.
• tessl schedule create and tessl task create honor TESSL_WORKSPACE when a run names an environment
• Attempting to install context from private repos will now fall back to SSH after attempting HTTPS

2026-10-14

#0.115.0

#New Features

• tessl launch models - List the agents and models cloud launches accept in a workspace, with the --agent spelling tessl launch skill takes and the default pair marked. Accepts --workspace and --json
• Run a schedule as one of its triggers - tessl schedule trigger --trigger <name> runs the schedule as that cron trigger, with its overrides

#Improvements

• Schedules belong to a project - tessl schedule create takes --project-id and otherwise uses the project linked to the current directory.
• Faster commands - Commands no longer spend about 0.7s checking the firing interval of each schedule in tessl.json. A schedule that fires more often than every five minutes is still rejected by tessl schedule validate, tessl doctor and on push, but other commands no longer warn about it
• Project linking explained - tessl project create, link and repair say when they change the name in tessl.json, show the old and new values, and ask you to commit the file
• Unconfigured repositories - tessl install in a repository with no tessl.json suggests running tessl init
• Install policy output - When install policy blocks a plugin, matching add and sync decisions are shown once under the evaluated registry version, and managed-project notes no longer repeat a security block's explanation
• Bundled code review plugin - tessl agent bundles tessl/code-review 0.4.0, which sets up Code Review through the Tessl Review GitHub App rather than a GitHub Actions workflow

#Bug Fixes

• tessl change verify --base <base> reads the old tessl.json at the merge-base of <base> and HEAD, so a branch behind its base no longer picks up the base's later verifier changes, and a partial clone no longer marks every verifier group as affected
• tessl project create offers only the workspaces of the resolved organization, and creates a new workspace in that organization
• After tessl login, the CLI suggests retrying an interrupted install only for the current repository, and ignores one recorded against an older tessl.json
• tessl install, tessl install --global and tessl init say "No plugins to install or update" when there is nothing to install

2026-10-09

#0.114.0

#New Features

• Region selection - --region us|eu, TESSL_REGION, or "region" in tessl.json chooses which Tessl region the CLI uses. tessl login stores the region you approved the sign-in in, so later commands use it without TESSL_ENV
• Repository settings - tessl code repo list, view and update manage each repository's Code Review settings, with --review-enabled, --review-trigger, --review-publication, --review-auto-approve and --review-cli-channel or --review-cli-version
• Launch cancellation - tessl launch cancel <id> requests cancellation of a launch run, and tessl launch list --status filters by pending, completed, failed or cancelled
• Linear access for environments - tessl environment create and update accept --linear-access none|read|write, so cloud launch runs can reach Linear through a run-scoped proxy without a personal token
• Verify without a login - tessl change verify runs without a Tessl login when both TESSL_LLM_BASE_URL and TESSL_LLM_API_KEY are set
• Verify finding locations - tessl change verify reports the lines a failed check refers to, in terminal output, --json, and GitHub annotations

#Improvements

• Setup links existing projects - tessl init and tessl project create link the checkout automatically when exactly one project matches the repository. tessl project create --new still creates another project
• Schedule sync validation - tessl schedule validate checks a linked project's schedules against Tessl and reports each failure by schedule key. Offline, it warns that the check was skipped
• Schedule branch default - A schedule that names no branch starts from the project's Loop default branch, or the repository's default branch, instead of main
• Launch status - tessl launch list, launch view and schedule runs show agent failed when the skill reports a failure and error when the platform fails. tessl launch skill --cloud --json reports a cancelled run as cancelled instead of failed
• Default verifier judge - tessl change verify uses openai/gpt-6-luna when --model is not set
• Faster verifier discovery - tessl change verify reads only the files matching each verifier group's glob, and skips nested .claude/worktrees and .codex/worktrees
• Longer pull request context - tessl code review reads pull request descriptions and comments of up to 16,000 characters, up from 4,000
• Exit alias - Interactive tessl agent sessions accept /exit as well as /quit

#Bug Fixes

• Long-running commands such as tessl eval run --wait refresh an expiring login instead of failing. If authentication is lost while waiting, the CLI says the eval continues on the server and how to check it
• tessl login no longer saves a session for an organization the region does not serve. It switches to the one organization the region serves, or lists them and suggests tessl login --loopback to choose
• A tessl.json that names its project without the workspace is treated as linked when exactly one project matches, so tessl eval run, tessl schedule list and install policy work in that repository
• tessl install stops without changing anything when Tessl cannot validate a policy-managed project, instead of applying policy without it
• tessl install evaluates install policy for projects with more than 200 dependencies
• tessl install reports a malformed tessl.json as a validation error instead of an unexpected failure

2026-10-01

#0.113.0

#New Features

• GitLab merge request reviews: tessl code review --pr <IID|URL> now reviews a GitLab merge request, including its notes and discussions, and selects the merge request automatically in GitLab CI pipelines. --publish comment posts a summary and inline discussions, and resolves or reopens them on later rounds. Authentication uses glab locally or a CI API token.
• Linked documents in code review: When a pull request asks the review to check an external link, such as a dependency's upgrade guide, the review fetches up to 5 requested links and every reviewing agent can read them. HTML pages are reduced to their text, and links the pull request does not ask about are never opened.
• Linked-source record: tessl code review lists which requested links were fetched and which were not, with the reason, in its terminal output and under linkedSources in --json
• Finding trigger events: tessl doctor accepts automation triggers on the Tessl finding.created event

#Improvements

• Check triggers: tessl doctor accepts automation triggers on GitHub check_suite and check_run events
• Minimal effort: Automation actions accept minimal as a reasoning effort
• Verify cost guidance: tessl change verify --help explains how judge calls spend credits and how to bound a run, and the --max-calls guard repeats it before a large run

2026-09-29

#0.112.0

#New Features

• Eval wait timeout - tessl eval run --wait-timeout sets how long to poll each run, in minutes or as a duration such as 90m or 2h, and turns on polling without --wait. The default wait is now 2 hours instead of 30 minutes. When the limit is reached, the CLI says the run is still in progress and prints the tessl eval view command to check it.
• Variants as JSON - tessl eval run --variant-json takes the variants to compare as inline JSON or a JSON file, with optional localPath fixtures uploaded from your machine. It can't be combined with --arms-json.

#Improvements

• Verifier help - tessl change verify --help shows exclude and content filters in its setup example, and links to the verifier docs and the tessl/tessl-verify plugin. tessl change --help names its verify and risk subcommands.

#Bug Fixes

• tessl agent retries a turn when the model connection drops with The socket connection was closed unexpectedly, instead of ending the turn with an error

#Deprecations

• tessl eval run --arms-json still works but prints a deprecation warning. Use --variant-json instead.

2026-09-24

#0.111.0

#New Features

• API key TTL - tessl api-key create --ttl <duration> sets a key to expire after a whole number of hours, days, weeks, months (30 days) or years (365 days), written as 3h, 2d, 4w, 6mo or 1y

#Improvements

• Deprecation notices on install - tessl install warns when a plugin you install is deprecated, shows the owner's reason, and prints the install command for the recommended replacement when one is set. The install still completes. The notice appears when you install a plugin without pinning a version.
• Full API key expiry - tessl api-key create prints the key's expiry as a full date and time in UTC instead of a bare date.

2026-09-23

#0.110.0

#New Features

• Jev verification - tessl change verify --model jev can use Jev for pass/fail decisions.
• Scoped skill search - tessl search accepts --org and --workspace to limit results to an organization or workspace the current user can access.
• Agent credit awareness - tessl agent shows session credit use and remaining allowance, with notices as allowance use rises.

#Improvements

• Verification dry runs - tessl change verify --dry-run is now more compact. Use --show-files to inspect matched files.
• Publication guidance - Plugin and skill publishing output distinguishes upload acceptance from publication approval and gives clearer steps when publication is blocked.
• Schedule diagnostics - Schedule validation warns when a linked workspace lacks an environment named by a schedule, and warns when a schedule names a multi-skill plugin without selecting a skill.
• Launch resume limit - A launch run can be resumed for seven days. The CLI help now states that limit.

#Bug Fixes

• Install sync - tessl install can sync dependencies when an unrelated schedule block is invalid; schedule validation still reports the problem.
• Verifier result validation - Verification rejects missing, duplicate, or renamed checklist results and refreshes invalid cached results before reporting a verdict.
• Global policy reconciliation - When the organization for a managed global install is ambiguous, the CLI stops before changing installed plugins, rules, or skills.
• Eval comparison values - Compact tessl eval compare output pairs standard deviation with the mean used for scores, cost, and duration.
• Fixture path handling - Local eval fixtures resolve symlinked parent directories correctly while retaining path boundary checks.

2026-09-15

#0.109.0

#New Features

• Tessl Code Review is out of beta - a completed review round costs a flat 50 credits, a review stops when the organization has no credits left, and the spend shows up in tessl org usage
• GitHub access on an environment - tessl environment create and tessl environment update take --github-access none|read|write, which sets the GitHub access granted to runs using that environment. tessl environment list and tessl environment view show it, and --env-file is no longer required when you pass --github-access
• Schedule capacity warnings - tessl schedule validate warns when a manifest declares more schedules than the workspace has active slots for. The command still exits 0 and reports valid: true, and --json output carries a new warnings array
• tessl schedule list - warns when the linked project's last schedule sync failed, with the time it failed and the reason

#Improvements

• @every schedule expressions - accept the full Temporal duration grammar rather than only whole minutes or hours
• Cron rejection messages - a full month or day name is corrected to its three-letter abbreviation, an out-of-range field names the field and its bounds, and a six-field expression is refused without parser detail

#Bug Fixes

• A cron that fires too often points at the field that makes it too frequent instead of the whole entry
• An invalid schedule key reports the full key rule against that key, rather than losing the message
• tessl workspace create asks which organization should own the workspace instead of failing when you belong to more than one, and an organization API key can now create a workspace

2026-09-11

#0.108.0

#New Features

• tessl eval run --arms-json - supply comparison arms as inline JSON or a JSON file path. Malformed arrays, more than ten arms, non-object entries, and missing or duplicate labels fail before the run starts
• Readable schedule formats in tessl.json - a schedule accepts raw cron, a closed set of readable shorthand strings, or a structured object, and schedule works as an alias for cron
• Agent signals from plugin hooks - skill activations, MCP tool activations and session availability are captured when hooks fire

#Improvements

• Dangling skill links in tessl doctor - a skill link whose target has moved is reported as a blocker with tessl install as the next action, in both the human output and --json. Renaming a project folder on Windows breaks every link at once, and doctor previously reported everything healthy
• Eval comparison tables - tessl eval run renders a compact arm comparison table before the per-arm scenario tables, with score, cost, duration, token and turn values in separate columns. In --full mode distributions render vertically and each scenario's criterion table follows its arm table
• One table style - every human-readable table shares a single borderless style with bold headers
• Code Review setup link - the embedded tessl/code-review plugin moves to 0.3.1, so installing the App from tessl agent lands on the organization's Code reviews page, which carries connect, repository selection and the two configuration questions in order
• tessl schedule list - the truncation notice is reworded

#Bug Fixes

• A PostToolUse hook accepts an MCP content array as tool_response, which several native adapters rejected before the hook script ran
• A Claude Code SessionStart model is preserved when the native payload supplies one, and Claude Code PostToolUse is no longer expected to carry a duration
• Recursive verification rejects a childManifests path that resolves outside the active project root
• Launch run live progress defaults to metrics.json, the artifact every recipe produces

2026-09-09

#0.107.0

#New Features

• Organization selection for Code Review - tessl code review accepts --org and TESSL_ORG. The selected organization drives model availability, credit preflight, key minting and billing attribution, and stays pinned to the key even if the account default changes later
• tessl outdated --global - check the plugins installed in ~/.tessl from any directory, with or without a local tessl.json
• tessl schedule list --all - page through every schedule instead of the first page. Without it, a notice now says how many are shown when more exist

#Improvements

• Reviewed code state in Code Review output - tessl code review names the selected head and comparison base, and warns when a committed selection leaves uncommitted local edits out of the review
• @tessl-reviewer - accepted as an alias for @tessl-code-review when triggering a review from a pull request comment
• Tessl Verify model default - tessl change verify and tessl change verify suggest-scopes now run on GPT-5.6 Luna. An explicit --model still takes priority
• Eval view statistics - tessl eval view builds its rollups and compact summaries from distributions, pairs spreads with means, labels solution cost as mean/solution, keeps the median in --full output, and marks a distribution that is incomplete

#Bug Fixes

• tessl code review --json emits one JSON document with nothing on stderr, so merged output streams parse cleanly through model retries and command failures
• tessl eval view shows why a scenario was excluded by the quality check rather than only that it was
• tessl eval run credit estimates count every arm, so a multi-arm run is no longer under-quoted before it starts
• tessl update reports every dependency whose version check failed, including private ones, instead of hiding them when another dependency is outdated
• tessl agent no longer leaves a stale status line frozen in the scrollback after a turn ends
• Filesystem, network and out-of-memory failures are reported as clear errors instead of surfacing as crashes

2026-09-08

#0.106.0

#New Features

• Desktop notifications - the agent rings the terminal bell and fires an OS notification when a turn finishes or a permission prompt appears while your terminal is in the background. Nothing fires while the terminal has focus, and print mode never notifies. Background detection relies on ANSI focus-event mode 1004, so a terminal that does not report focus changes is treated as focused and stays quiet. Turn it off with tessl config set notifications false
• Every agent run ends with a summary - an interactive tessl agent run now closes with one block: the files it recorded writing or editing, and one suggested next action, offered as prompt ghost text you accept with Tab. The file list covers edits made through the edit and write tools, so a run that reports none is saying it recorded no such edit, not that nothing on disk changed
• --no-wait - tessl eval run, tessl review run (quality and security) and tessl scenario generate take --wait and --no-wait. With --no-wait each submits the work and returns its run id, or ids for an eval run that submits several, without polling, so a CI job can start a run and collect the result later with the matching view command. It is rejected alongside a gate that needs the run to finish: --threshold on a quality review, --fail-on on a security review
• Multi-arm eval runs - tessl eval view detects a run with custom arms and renders one section per arm, each with a rollup line and per-scenario rows carrying score, cost per solution, turns, tokens and agent time. --full expands every stats cell to the full distribution: min, q1, median, mean, q3, max and sd. The classic baseline and with-context pair keeps its existing layout, now with compact stats lines. Arms mode never shows lift

#Improvements

• The GitHub App is the supported way to install Code Review - the docs and every CLI surface that pointed at the GitHub Action now point at the Tessl Review GitHub App, and the Action is marked unsupported. The tutorial is rewritten around installing the App, enabling a repository, trigger policy, approval, the three comment intents and the .tessl-code-review.yml reference; running Code Review in your own CI becomes a sub-page of it. The CLI embeds tessl/code-review 0.3.0
• Deleting your last workspace deletes it - the old behavior removed the workspace and immediately re-created an empty copy in its place, then explained across six messages that it had been kept. An organization may now hold no workspaces
• A warning before a workspace name is lost - deleting a workspace that has published a public plugin permanently reserves its name, and an interactive tessl workspace delete now says so while you can still change course, along with pointing out that moving a workspace between organizations is a support request rather than a delete and recreate. A scripted deletion does not show the preview
• Windows skill installs no longer need Developer Mode - tessl install and tessl init create a directory junction on Windows instead of a symbolic link, which a standard user can create without elevation. Installation had been failing outright on devices whose policy blocks Developer Mode
• tessl skill review is removed. It has printed a deprecation banner since June and the migration guide's removal date passed in July. Invoking it exits non-zero with "No review ran" and names tessl review run and tessl review fix as its replacements, and it still parses every flag the old command took, so a pipeline reaches that message rather than a flag-parse error. The flag-by-flag mapping is in Migrate from skill review

#Bug Fixes

• Ctrl+C works while a spinner is running. tessl code review made this visible because its live lens board keeps one spinner up for the whole run, so the command could not be interrupted at all, but every spinner in the CLI was affected whenever nothing had read stdin before the spinner started
• A tool call that fails and is recovered from on the next turn renders on the neutral gray background rather than in a red error box. The error output is still shown. Stack traces from background dependencies no longer reach the terminal
• The permission dialog's options each state their own effect, ordered narrowest to broadest: allow once, allow this pattern for the session, block once, and block with an optional reason for the agent
• Every interactive session has exactly one prompt ghost. Until the conversation starts it offers the splash question, and after every completed turn it offers the next thing to type, in every session rather than only one started with /skill:<name>
• tessl scenario generate against a repository now fails when a generation completes with no scenarios, matching what the plugin path already did. The exit code carries that verdict only on an invocation that waits for the run, so a CI pipeline gating on it needs --wait, since the command otherwise submits and returns before the outcome is known. The human output counts scenarios rather than generations

2026-09-03

#0.105.0

#New Features

• Live lens board - tessl code review shows a per-lens progress board on stderr while it runs, marking each lens as queued, reviewing, a finding count, failed, or skipped before the reconciling step; --no-progress hides it and --json implies it
• Lift as the eval headline - tessl eval run prints a Lift: line under each completed comparison run, and tessl eval view leads its summary and each scenario with the lift in percentage points, noting how many unscored results were excluded
• Organization-level project discovery - tessl project list, tessl project link, and tessl project create discover projects across your whole organization, with --org (or TESSL_ORG) to pick the organization, --workspace for single-workspace behavior, and tessl project link --path to match a package project by its repository subpath

#Improvements

• One Tessl Code Review plugin - the CLI embeds tessl/code-review 0.2.0, which now carries the setup, lens-authoring, and review-response skills alongside the four lenses; the default lenses are repinned to 0.2.0 with unchanged content, and a project that still declares tessl/code-review-setup or tessl/code-review-lens-creator in tessl.json keeps using it until it switches
• Code review at splash - the "Review code changes" onboarding workflow installs tessl/code-review, sets up review on every pull request, and suggests one custom lens the default lenses would miss, without writing it
• Grouped review summary - tessl code review ends with a versioned header, a verdict line with must-fix and suggestion counts, findings grouped into must fix and suggestions sorted by severity, and the exact publish command; under --json the same summary goes to stderr so CI logs carry it
• Eval reuse made visible - tessl eval run and tessl eval view print how many results were reused from previous runs beside the cost line, so a fast, cheap run no longer reads as stale
• tessl project list --json reports the organization - the output carries a top-level org instead of workspace unless --workspace or TESSL_WORKSPACE pins one

#Bug Fixes

• tessl code review against a caller-supplied LLM endpoint with a model-scoped key no longer fails during reconciliation, because the pinned supervisor model is sent by its bare name
• tessl eval view floors score percentages instead of rounding them, so the table, averages, and lift agree
• The splash screen's workflow picker moves one step per arrow key press on Kitty-protocol terminals such as Ghostty, Kitty, and WezTerm, where the key release was counted as a second press; a held arrow still scrolls smoothly

2026-09-02

#0.104.0

#New Features

• Eval cost shown before you spend - tessl eval run prints an estimated cost line and your org's daily eval usage before the run starts, including with -y
• Eval cost in machine output - tessl eval run --json gains estimatedCredits and quota, and tessl eval view --json gains meta.credits for the settled run cost

#Improvements

• Review action beside severity - every published Code Review finding now shows Requires changes or Advisory next to its severity, along with why an advisory finding is advisory: low confidence, pre-existing, unclear relationship, or the review mode
• Run cost reported in CI - the post-run credit cost line is no longer suppressed on non-interactive streams, so CI and piped runs see the spend

#Bug Fixes

• tessl api --input - and -F key=@- now read stdin from a redirected file or heredoc, not just a pipe; both refuse an empty stdin or a bare terminal with an error instead of sending no body

2026-09-01

#0.103.0

#New Features

• Scheduled runs - tessl schedule is now available. You can use it to automatically run skills on a periodic schedule.
• Environment management - tessl environment can be used to manage environments for tessl schedule and tessl launch skill --cloud
• Antigravity support - first-class support for the Antigravity CLI (agy) and the Antigravity IDE, alongside the existing Gemini integration
• tessl eval run --wait - polls the run to completion and exits non-zero when all runs fail; --json always submits without waiting, and a run that outlives the poll window exits 0
• tessl api-key delete --yes - skips the confirmation prompt, with -y as the short form

#Improvements

• Agent step lines - each tool call in an interactive tessl agent session renders as one status line naming the step and its result, with ctrl+o to expand the full output
• Expand-all hint - the agent header now advertises the collapse and expand-all shortcut, resolved from the effective keybinding
• Activity footer - the footer names the running step, shows an elapsed timer, and keeps the known-total count, then leaves a one-line result carrying the total duration
• Permission prompts - the command, target, or path appears on its own indented line, long commands are middle-truncated so the tail stays visible, and bash prompts carry a Why line when the agent states its intent
• Code Review finding assessment - the supervisor now assesses likelihood, confidence, and relationship to the change alongside severity, and a deterministic Strict, Standard, or Relaxed policy decides what blocks
• Code Review severity table - rows are ordered Critical, Major, Minor, then Nit rather than by insertion order
• Code Review round notice - rounds after the first carry a collapsed "About this review" block noting the findings are AI-generated, with a link to the guidance
• Code Review thread links - a finding carried forward now links the thread that is carrying it
• Skill review phases - the bundled review-skills-in-repository skill splits into a Discover phase that reports and stops, and a Review phase entered only when you opt in
• Skill review scope - after discovery the skill offers one skill, a prioritized batch, or every discovered skill, each with an expected duration and credit cost, replacing a silent 20-skill cap
• Skill review output - the review leads with a ranked list of every reviewed skill, one line each, with per-skill detail on request and coverage counts below it
• Skill discovery - discovery runs as a deterministic bundled script rather than an agent file-system walk
• Clearer command briefs - tessl project, tessl agent, tessl outdated, and tessl update now describe what they do
• Clearer skill name errors - an invalid skill name names the specific problem and suggests a corrected name
• Eval default model notice - the notice no longer names a model and points at tessl eval run --list-agents, since the default is declared per region
• Project mismatch - tessl install blocks with the relink command when the backend already has a project for this directory, and tessl init warns before creating a second one

#Bug Fixes

• tessl eval view exits 1 for a failed run, and tessl eval run aborts with exit 1 when the scenario count exceeds the guard
• npm install tessl@X installs version X instead of whatever the release channel currently advertises
• The agent session no longer clears scrollback and snaps to the top while streaming, so a permission prompt stays reachable
• Code Review keeps successful lens results when a sibling lens fails, and states the reduced coverage instead of discarding the whole review
• Code Review republishes a moved finding at its current anchor and links the earlier occurrence, instead of leaving the only comment on stale code
• Code Review includes verified optional findings in first and explicitly requested full reviews
• tessl workspace delete distinguishes a deleted workspace from one emptied in place, and names the plugins to remove when they block deletion
• tessl scenario generate no longer blames plugin content or the quality verifier for a zero-scenario result, and reports a timeout distinctly from a failure
• tessl scenario download defaults to the plugin root when run inside a plugin
• Fixing path handling bugs in tessl install and tessl init on Windows
• The agent onboarding editor keeps a custom editor factory and honors the Kitty keyboard protocol
• An ambient WORKOS_CLIENT_ID no longer overrides CLI auth

2026-08-28

#0.102.0

#New Features

• tessl eval lint - the scenario lint command now shows in tessl eval --help, and tessl scenario download prints a hint that downloaded scenario files can be edited by hand and validated with it
• Launch activity for pending runs - tessl launch skill --cloud --wait, tessl launch view, and tessl launch list now show a pending run's phase and last progress time; retrying and recovering read distinctly and are never shown as failed

#Improvements

• Adaptive review effort - tessl code review now defaults to adaptive effort, picking the right depth for each review; --effort with low, medium, or high stays fixed
• Code Review finding quality - findings are checked more rigorously before they are surfaced, and severity now better reflects the real consequence of each finding
• MCP server status - configured MCP servers now read as ready and connect automatically on first tool use; the footer, status output, and MCP panel share one state model with clearer sign-in guidance
• Plugin terminology - user-facing copy now consistently says plugin and plugins; the migrate commands for legacy tile content are unchanged

#Bug Fixes

• Code Review now replies on an existing review thread when a re-review changes its conclusion, instead of leaving its earlier answer as the last word while the check stays red
• tessl login now lets a user who belongs to more than one organization choose which one to sign in to on the approve page, instead of failing for members of an SSO-enforced organization with a second membership
• Spinners are suppressed when stdout is not a TTY, keeping piped output, --json mode, and agent contexts clean
• tessl cli update now shows a clear error pointing at --channel when a release channel name is passed as --target
• Eval runs that outlive the CLI's poll window are now reported as timed out instead of failed, so a run that completes server-side no longer triggers a false "All eval runs failed"
• The workflow hint in tessl agent now steps aside while another UI such as the settings menu owns the prompt row, and no longer appears on --continue or --resume

2026-08-27

#0.101.0

#New Features

• tessl login - loopback is now the default transport. Plain tessl login uses the loopback authorization-code and PKCE flow, and the device grant stays available with --device.
• Local GitHub App minting - when TESSL_GITHUB_APP_ID and TESSL_GITHUB_APP_PRIVATE_KEY are both set, tessl github token, tessl github setup-auth and the auto-refresh sidecar mint installation tokens directly as that App, so its own bot identity lands on commits and pull requests. With either variable absent, behavior is unchanged.
• Guided first run - a new signed-in user is handed off from the web onboarding into tessl agent with the review-skills workflow, with in-editor hints for the next action.
• Context and Findings - the embedded CLI skill now works in terms of Context and Findings instead of a skill inventory, and points at the current per-workspace Context and per-org Findings endpoints.

#Improvements

• tessl code review - a finding states what is broken and what a fix has to achieve instead of prescribing a specific fix, and a review no longer blocks on code outside the diff.
• A deliberate, explained reversal of an earlier fix is read as a decline rather than a regression, so the same concern is not raised again.
• A finding is reposted on a silenced thread only when it requests changes, so an optional suggestion is no longer republished every round.
• Local-mint failures show GitHub's own error message without --verbose, and report the measured local-versus-GitHub clock difference when the two disagree.
• The --model and --scorer-model briefs on tessl eval run show current model names.

#Bug Fixes

• tessl login reports the reason for a WorkOS authentication challenge, such as organization selection or email verification, instead of a bare status code.
• tessl login --loopback always runs the loopback transport instead of silently rerouting to the device grant on a host that looks remote. An automatic fallback prints why it happened and which flag overrides it.
• Skill symlinks resolve the physical parent directory before computing their target, so links no longer dangle when an intermediate directory is itself a symlink.
• tessl launch skill no longer reports a watch timeout as a launch failure. It stops watching, exits 0, and prints a tessl launch view <id> command.
• tessl eval run refuses a run started from inside a plugin's evals/ directory before any credits are spent, rather than silently running baseline only with no plugin context.
• Onboarding hints release the arrow keys when another component, such as the model picker, takes focus.
• The [PostHog] Timeout while shutting down PostHog warning is no longer printed on exit.

2026-08-26

#0.100.0

#New Features

• tessl project list - lists the projects in a workspace with each one's name, ID and source. The workspace is resolved from --workspace, TESSL_WORKSPACE, the project linked in tessl.json, or a picker. --json emits a structured list and never prompts
• A machine-readable verify report - tessl change verify --report-json <path> writes its JSON report to a file, so CI can consume verifier findings without a second run. Stdout is untouched and --github annotations still work alongside it. The report now also carries each verifier's stable id, in both --json and --report-json
• Cloud launch progress - tessl launch skill --cloud --wait shows live progress on the spinner while the run is in flight, instead of a static message. --json and --no-wait are unaffected

#Improvements

• Working with this review - the first Tessl Code Review on a pull request now publishes a collapsed section telling the reader that the findings are AI-generated and not authoritative, which answer to give on each thread (addressed, refuted, or declined), that reviews are incremental and what a later round therefore withholds, and that lenses are configurable and glob-scoped. It is published on the first review only, so a later round does not push the findings further down. tessl code review run -h says the same things
• Settled threads close themselves - a Tessl Code Review round now resolves the threads whose finding the current head fixes, after replying on them, so what stays open on a pull request is what still needs someone. A refuted or declined finding's thread stays open for the author to close
• One closing instruction on every round - every published review round now closes with the same line: mention @tessl-code-review to run another review

#Bug Fixes

• Repository flags accept a GitHub URL - tessl github setup-auth --repo and tessl github token --repo accept a full GitHub URL as well as an owner/name slug. HTTPS, scheme-less github.com/owner/name, and SSH forms all resolve to the same repository, and a non-GitHub host is rejected with the value echoed back
• A full review says it was full - a full Tessl Code Review no longer closes by claiming it narrowed itself. A full round now states that the whole change was read and every new finding raised whatever its severity, and keeps the contrast with an ordinary round
• A live finding is not left collapsed - a re-review that concludes an earlier finding still applies now reopens that finding's resolved thread, so the update is not left where tooling and readers skip it. A thread whose finding was answered as addressed, refuted, or declined stays resolved
• A finding reaches its reader when a thread cannot reopen - the finding is published as its own comment instead of being left where no reader reaches it. A GitHub Actions token may reply to a review thread but may not unresolve one, so the review now asks per thread rather than assuming
• Mentions are matched case-insensitively - a mention of @tessl-code-review that differs only in case, such as @Tessl-Code-Review approve, is now read as a mention. An approval request spelled that way was previously answered with an ordinary review
• Verify reaches allowlisted dot-directories - tessl change verify discovers files in repository-root dot-directories that a verifier's include globs name explicitly, so an include such as .github/code-review/*.mjs matches instead of silently matching nothing. Only allowlisted repository-tooling directories are opened, and only at the repository root

2026-08-24

#0.99.0

#New Features

• Named approver logins - tessl code review --approver <login> names comment authors permitted to request an approval, beyond the repository owners, members and collaborators who always may. The flag is repeatable, logins are compared case-insensitively, and a bot login is matched exactly as the event payload spells it, [bot] suffix included. GitHub reports every app comment with association NONE, so before this no bot could pass the guard
• A refused approval request no longer runs a review - asking to approve without permission settles as skipped in under a second, rather than spending a full review to answer the opposite of what was asked. The message leads with what was refused and names the route back: approving is limited to owners, members and collaborators, plus any login passed with --approver
• Full review requests - a commenter who may approve can ask for a full review instead of another round. Its specialists then see the whole change rather than only the commits since the last reviewed head, and a low-severity finding no earlier round raised is raised. Prior findings still reconcile on their existing threads. A request from an author the repository does not trust to approve runs the ordinary round rather than being refused
• A change-requirement threshold in Code Review profiles - a repository YAML profile sets requestChangesAt, the severity at which a finding starts requesting changes, so a team adopting Code Review on a large existing codebase can gate on criticals alone while a repository with a strict main branch holds minors on every round. It is a profile key rather than a flag, so no caller can weaken the gate without changing the profile the repository committed. The default is now major on every round, which means a minor no longer requests changes on a first review
• Machine-readable output on list and view commands - --json on tessl whoami, tessl config get, tessl config list, tessl org list, tessl plugin info, tessl workspace list and tessl workspace list-members prints one JSON document. Every --json command now writes one pretty-printed document to stdout and routes hints, progress, warnings and errors to stderr, so stdout stays parseable
• Sign in with a local browser redirect - tessl login --loopback signs in with a loopback authorization code and PKCE, so the browser hands the session back to the terminal without a code to type. The device code flow stays the default and is unchanged, and --loopback falls back to it when the browser is not on this machine or the loopback port cannot bind
• Find work to automate - a fourth onboarding workflow in tessl agent looks for what you do repeatedly that could run itself. It returns at most five ranked candidates, each with its evidence, its expected value, a suggested trigger and an implementation direction, and expands one on request

#Improvements

• Whoami describes the API key - tessl whoami in a session authenticated with an API key now shows the current key in use (masked), its scope, its id, and the organization and workspace it belongs to
• Findings below the threshold on a later round - a re-review no longer raises a fresh finding below the change-requirement threshold, and a finding an earlier round raised stays on its thread. An approving review that carries findings now states how many of them are optional, so open findings under an approved verdict are not read as blockers
• Guidance routing - broad questions about Code Review, Risk and Verify reach the harness design skill, while an explicit request to author a review lens still reaches the lens creator
• Code Review setup knows the new profile keys - the bundled tessl/code-review-setup moves to 0.3.0, so a project that does not declare the plugin itself gets a setup skill that covers the change-requirement threshold and who may ask the reviewer to approve
• A rejected CLI version says how to upgrade - when the backend refuses a CLI as too old, the CLI prints the reason it was given along with how to upgrade, rather than a generic failure. A request that never reaches the backend at all is now reported as a network failure rather than as an API rejection
• Clearer guidance outside a project - a command run where no Tessl project can be found says to run tessl init from the project root, instead of describing what a project is

#Bug Fixes

• Spinner frames stay out of JSON output - tessl review fix and tessl review run quality stop the spinner before printing the --json status payload on the timeout, failed and cancelled paths
• Code Review resolves the repository's default branch - the base is read from origin/HEAD, falling back to a probe for origin/main then origin/master, instead of assuming origin/main. A repository whose trunk is named something else is told to pass --base rather than failing on a ref that does not exist, and an explicit --base still wins outright
• An empty change is skipped rather than approved - a local review of a clean, up-to-date checkout returns a skipped result carrying No changes to review. before any model call, instead of running every lens and publishing Changes approved over a change that does not exist
• Eval runs report what happened - tessl eval run retries a run that is momentarily missing from the poll endpoint just after it was created, rather than treating it as terminal, so completed runs no longer report All eval runs failed. When every run really does fail the command exits non-zero, so a pipeline wrapping it sees the failure

2026-08-20

#0.98.0

#New Features

• Publish a Code Review from the CLI - tessl code review --publish <comment|verdict> posts the completed review to the pull request it reviewed as one native GitHub review: a summary body, an inline comment on each finding GitHub can anchor to a changed line, and a reply on each earlier finding's thread the round reconciled. comment states no position on the change, while verdict carries the review's own conclusion, approving when nothing requires changes and requesting changes otherwise. With --json, the result carries a publication receipt naming what was published
• Reasoning effort is supported configuration - --effort low, medium or high sets how hard each review lens thinks. A YAML profile can set effort for every lens, and any single lens can set its own to think harder or less hard than the rest, so depth is spent where it pays. A lens without one uses the profile's, with neither the model applies its own default, and --effort overrides both
• Ignore globs in Code Review profiles - a YAML profile takes a top-level ignore list of repository-relative POSIX globs bounding which paths any lens reviews, so generated code, lockfiles, snapshots and vendored directories are excluded once instead of negated inside every lens
• An inline workflow hint replaces the onboarding splash - tessl agent opens with a few lines above the prompt rather than a full-screen welcome. Either arrow key opens the list, selecting a workflow fills the prompt with it ready to send, and /getting-started reopens the list
• Review my skills - a new onboarding workflow that reviews every skill in the repository. It discovers them, lints them deterministically, reviews each against seven dimensions, and returns a ranked plan tied to evidence. It changes no files, and offers tessl review fix as an opt-in follow-up
• Learn from PR feedback needs no input - the workflow defaults to the last two weeks instead of asking for a pull request number or a time period, states up front that it changes no files, and returns a ranked, capped set of improvements with its coverage limits named
• Workspace and organization environment variables - TESSL_WORKSPACE and TESSL_ORG set the workspace and organization for commands that take --workspace and --org, so a script does not repeat the flag on every call. An explicit flag always wins, and a project .env file cannot set either one
• Global installs follow organization policy - globally installed managed requirements converge on the organization's install policy whenever the CLI contacts the backend, not only on an explicit tessl install -g. Security thresholds, minimum release age and source restrictions all still apply, and only policy-owned content is removed

#Improvements

• Clearer failure messages - installing over Git, self-update, sign-in and eval artifact uploads report failures written for the reader, rather than passing transport wording such as HTTP 404 through to the terminal
• Help reads for its audience - tessl --help prints plain text for a person, with usage, grouped commands and flags. The Markdown reference is served only to recognized agent callers, and covers the top level, a group and a leaf alike
• Code Review setup - the bundled tessl/code-review-setup moves to 0.2.0. The setup skill documents the profile ignore list, and the workflows it writes subscribe to pull request review comments, so a mention left on the diff starts a review. Who may request one moves to an allowed-associations input rather than a condition the workflow implements itself, and the uses: line references the Action's major tag instead of a commit resolved at setup time, so a repository picks up fixes without editing its workflow

#Bug Fixes

• Conceded findings settle - a finding the author agrees is real but defers to tracked follow-up work now reconciles as declined, instead of resurfacing as remaining on every later round
• Verify separates errors from failures - tessl change verify reports a target it could not judge, such as a base ref it cannot resolve, as a distinct error outcome rather than a verifier failure. Those targets are listed under "Could Not Be Verified", left out of the failure breakdown, annotated as notices, and still exit non-zero
• Launch reports the agent's own outcome - tessl launch skill --cloud --wait exits non-zero when the workflow completes but the agent reported a failure, and tessl launch view renders the agent-reported outcome for completed runs

2026-08-18

#0.97.0

#New Features

• Repository YAML profiles - tessl code review --profile <path> now accepts a repository YAML profile as well as the built-in standard profile. A profile routes lenses with ordered repository-relative globs, and each specialist's prompt and read-only diff tools are scoped to the paths its lens matched. The path must end in .yml or .yaml
• Approval commands - a trusted repository owner, organization member, or collaborator can mention @tessl-code-review with an explicit instruction to approve the pull request. Code Review returns an approved result without running specialist lenses. Ordinary review requests, ambiguous wording, and classification failures continue through a normal review
• Agent-facing help - a bare tessl, or an explicit -h, from a recognized agent now prints a Markdown description of Tessl and how to drive it, generated from the command registry. Claude Code, Cursor, Codex, and Gemini are detected without configuration
• Readiness reporting - tessl doctor --json and the MCP status tool both return a readiness field with a state, machine-readable blockers, and an ordered nextActions list. One shared function computes the field so the two surfaces return consistent guidance. Existing fields and doctor terminal output are unchanged
• API key listings as JSON - tessl api-key list --json prints a machine-readable listing, walks every page so a truncated result cannot appear complete, and includes each key's role
• API key expiry notice - tessl api-key create prints the key's expiry date and days remaining to stderr. A key piped directly into a secret store still reports its 30-day lifetime while stdout stays clean
• rotate-ci-key - a new bundled skill reads your pipeline to find which secret supplies TESSL_TOKEN, selects a rotation tier based on its access, presents a plan naming the source of each fact, and waits for approval before changing a live credential
• Antigravity skills - a global install now also symlinks skills into ~/.gemini/config/skills/, where Antigravity, Antigravity IDE, and Antigravity CLI look. The existing .agents/skills path is unchanged

#Improvements

• Lens limit - Code Review accepts up to 8 lenses, up from 5. The lenses active for the current change are resolved before the limit applies, so lenses whose globs match nothing do not count against it
• Error routing - 9 more precondition failures name what was found and return the command that resolves it. This covers linting a directory with a SKILL.md but no manifest, an unlinked tessl.json, an invalid package name, an empty scenario directory, a missing fix-run bundle, and the deprecated skill review path. Raw HTTP statuses no longer reach the user
• Setup handoff - tessl init now wires agent context (RULES.md and AGENTS.md) even when no plugins are installed, so an agent has context before anything is installed. tessl login resumes the interrupted setup step when there is one, and otherwise points to the first unblocked step. Every MCP tool that requires authentication now says so in its description
• Rejected key guidance - a rejected API key now reports where the detected CI provider keeps pipeline secrets and which of its CLIs edits them, across 12 providers with a generic fallback. On GitHub Actions, the failure is attached to the workflow file as an annotation and appended to the run summary. Outside CI, the guidance points to a TESSL_TOKEN that may be shadowing a healthy session and should be unset instead of rotated
• tessl inventory import scan scope - scan scope is read from the integration scan policy configured on the organization integration page and enforced when data is persisted. The command no longer fetches a server-side exclusion list to pre-filter the scan. Only --ignore-repo narrows a run locally
• Lens authoring - the bundled tessl/code-review-lens-creator moves to 0.1.1. The lens-authoring skill now directs its reference reads explicitly, and its evaluation tasks no longer prescribe the answer they score
• Code Review setup - the bundled tessl/code-review-setup moves to 0.1.5. The setup skill now covers YAML profiles for CLI and Action runs, including lens routing, path scoping, the lens limit, local reference safety, and what happens when no lens matches. It also configures the Action with a repository-relative profile path

#Bug Fixes

• Login link - the URL printed when the CLI cannot open a browser now always carries its verification URI. Production builds previously omitted it, leaving SSH and development-container logins with a link the capture page could not use
• MCP shutdown - Ctrl-C no longer crashes the MCP server, and its capability warning is gone
• Supervision context - a live review refreshes the pull-request conversation after the specialists finish and immediately before supervision, so a reply posted while the review was running is taken into account. The run fails closed if the refresh fails or the base or head moves during the review
• Reviews with no matching lens - a review whose profile matches none of the changed paths returns an explicit skipped result without running a model, instead of completing in a way that reads as approval
• Truncation reporting - prompt truncation is reported whenever any specialist lens truncates, rather than depending on which parallel lens finished last
• Review cancellation - the every-commit workflow template no longer lets a comment on a pull request under review cancel the review in progress, and a push to a draft can no longer cancel a requested review. Cancellation is limited to a non-draft synchronize event. Workflows generated before this release keep their original template until regenerated

2026-08-12

#0.96.0

#New Features

• Tessl Code Review (Beta) - tessl code review now appears in the CLI help and runs a multi-lens review over a code change. With no flags it auto-selects the current pull request in a supported CI environment, or otherwise the complete local change; --pr <ref> reviews a pull request, and --base / --head review an explicit Git range with no pull request required
• Code Review setup through Tessl Agent - the tessl/code-review-setup plugin now ships embedded in the CLI, so asking Tessl Agent to set up Code Review installs the review workflow directly; the embedded guidance for the older change review flow was removed
• Lens authoring through Tessl Agent - the tessl/code-review-lens-creator plugin also ships embedded, so asking Tessl Agent to create a review lens scaffolds one without an install

#Improvements

• Change review hidden from help - tessl change review no longer appears in help output; Tessl Code Review supersedes it. The command keeps working for existing callers
• First run points at setup - the first CLI invocation shows a one-time hint pointing at tessl init, and the install scripts hand off with the same guidance. The hint is skipped in CI and on invocations requesting --json
• Error messages direct the user to the command that resolves them - user-facing errors now route to the command that fixes the reported state, instead of only describing the failure. Additionally, every missing-argument error states which argument was missing
• Credit refusals are legible mid-run - when credits run out in the middle of a run, commands surface a clear credit message across CLI and MCP surfaces instead of a generic failure
• Skill review MCP tools state their scope - the review_run, review_fix, and review_view MCP tool descriptions now say they review skills, and point to tessl code review for reviewing code changes
• MCP tool descriptions name neighboring workflows - the status MCP tool description names the adjacent workflows to run next, so an agent picks the following tool without guessing
• Skill publish validation - publish validation reports every independently detectable problem in one pass, and skill validation messages use Tessl terms instead of raw schema output

#Bug Fixes

• Install with a broken project link - tessl install no longer fails when tessl.json links to a project that cannot be resolved, and errors in that state no longer suggest creating a new project
• Organization selection in workspace create - tessl workspace create resolves the owning organization before creating the workspace: the new --org flag takes an organization name or id, and membership in several organizations prompts for a choice
• Scenario generation from the current directory - the source argument to tessl scenario generate is now optional and defaults to the current working directory
• API token identity - tessl doctor, tessl init, and tessl mcp start no longer print a fabricated api-key@tessl.io email when authenticated with a TESSL_TOKEN API key; they identify the session by the masked token instead
• Corrupt tessl.json errors name the file - a JSON parse failure while reading tessl.json now names tessl.json in the error message
• Live documentation links - tessl skill new and tessl plugin new print working docs.tessl.io links in their next-step guidance
• No memory writes on read-only turns - Tessl Agent no longer creates or updates .tessl/memory files during turns that make no changes

2026-08-07

#0.95.0

#New Features

• Delete an API key by name - tessl api-key delete <name-or-id> accepts either the key name or its ID, matching the pattern already used elsewhere. Interactive selection still works when no argument is given
• Skill removal guidance - asking to unpublish, archive or delete a skill now explains that skills are published inside a plugin and lists the three real removal paths, and tessl skill --help carries the same tip
• Uncollected file warnings in review - tessl review run and tessl review fix now name the skill files that were not sent to the reviewer, so content that scored low because it never reached the judge is visible
• Out-of-workspace path warnings - tessl scenario view flags generated scenarios whose task or criteria reference a home-directory or absolute path, which can never score above zero because the grader only sees the workspace

#Improvements

• Credit warnings on stderr - credit warnings print to stderr instead of stdout, so they no longer sit inside a command's own output. TESSL_CREDIT_WARNING_STREAM accepts off, stderr or stdout to override the choice
• Upgrade path in the agent credit footer - the out-of-credits and over-limit states now point at /usage, where the balance and the upgrade link live
• Pricing links - credit and upgrade calls to action point at /pricing
• Credit checks on the MCP surface - the MCP server now runs the same credit preflight as the CLI, closing a path where a blocked organization could still spend credits
• Score composition in review output - review results show how a score was composed, per judge, rather than only a final number
• Organization default model for agent sessions - tessl agent picks up an organization's configured session model when no --model flag and no persisted default are set
• Review hands off to fix - after a quality review, tessl review run points at tessl review fix as the next step when the review found something to act on. The MCP review_view tool carries the same signal as a nextStep field
• Fix runs stop at a reachable score - the default fix threshold drops from 100 to 90, so a run can finish early instead of burning every improve iteration on a target a review almost never hits. The MCP surface no longer holds its own defaults and reports the config the server resolved

#Bug Fixes

• Review scores render against the rubric's own scale instead of a hardcoded denominator of 3, so a 1 to 5 rubric no longer reports the wrong total
• tessl skill review --json carries the deprecation notice in its payload, and the post-apply hint points at tessl review run instead of the deprecated command
• A logged-out global install of registry content is refused when the global context is managed by an organization policy, closing a path where logging out skipped the organization's managed requirements
• tessl install -g labels the resolved policy scope org: instead of mislabeling it workspace:
• Auto permission mode no longer fails at startup for organizations whose model catalog omits gpt-5.4-mini. The classifier falls back to the resolved session model
• Reasoning models in the gpt-5.6 family route through the OpenAI Responses API, fixing a gateway 400 when a request carried both a function tool and a reasoning effort
• Skill inventory upload accepts internal as a repository visibility, so GitHub Enterprise organization-internal repositories are no longer rejected

2026-07-31

#0.94.0

#New Features

• Credit usage meter - tessl org usage shows your organization's plan, credits used against the allowance, and when the current window started. Pass --org to name an organization, --json for the raw credit account
• /usage and /cost in the agent - /usage shows the same meter without leaving the session, /cost shows what the session has cost in credits with the token counts underneath, and a credits: N left segment on the footer refreshes each turn
• Commands report what a run cost - every command that waits for a run prints the credits it consumed once billing settles, and ... view commands report the cost of the run they inspect. The figure comes from the server, so it matches what you are billed
• Credit guardrails on billable commands - a one-line notice when you are over your allowance but still allowed to proceed, and a hard stop with an upgrade link when you are not. Both are suppressed under --json so structured output stays parseable
• Scoped reviews - new repeatable --path flag on tessl change review restricts the review to files matching a pathspec

#Improvements

• Security grades read Tessl's own levels - tessl search, tessl tile info and the security review output now render the Tessl security level and per-finding severity. A clean result reads "Passed", an unscanned one reads "N/A", low-severity findings group into their own section below higher severities, and an unrecognized level renders as-is instead of failing. --json keeps the previous fields alongside the new ones
• Eval fixtures are bounded to the run - tessl eval run and publish-time evals now refuse scenario fixtures and packed symlinks that resolve outside the run's repository or scenarios boundary, so an eval can no longer archive files from elsewhere on the machine. tessl eval lint reports the same. Out-of-boundary directory fixtures and symlinks are opt-in via tessl eval run --allow-unsafe-fixture-paths, or --allow-unsafe-eval-fixture-paths on tessl skill publish and tessl plugin publish. A scenario include must always resolve inside its own scenario, with no override
• A project's .env can no longer redirect the CLI's own state - TESSL_GLOBAL_DIR, TESSL_BINARY_BASE_URL, TESSL_AUTO_UPDATE_INTERVAL_MINUTES and TESSL_DISABLE_UPDATE are ignored when they come from a .env file in the working directory. Values set in your shell are unaffected
• Workspace resolved from a plugin manifest - working inside a plugin directory that has .tessl-plugin/plugin.json but no tessl.json no longer prompts for --workspace; the workspace is read from the plugin name
• Declared skills scanned as real files - tessl inventory import now scans skills declared in tile.json / plugin.json the same way it scans any other SKILL.md in the repository, so they carry their body and frontmatter instead of being recorded as bare references. Their names now come from frontmatter (or the containing directory) rather than being prefixed with the plugin name, and a declared skill whose file is missing from the scanned tree is no longer recorded at all
• workspace_list reports what you can do - the MCP tool now returns allowedActions per workspace, so a caller can filter to the workspaces it has permission to act on before trying

#Bug Fixes

• Failing commands print one error line. Previously the same message appeared twice, with two different cross glyphs
• tessl change verify --recursive --github now emits GitHub Actions annotations for every root. Warning-level findings from a recursive run previously only appeared in the raw job log
• Out-of-credits and other expected failures from the review_run, review_fix and uninstall MCP tools are reported as errors rather than crashes
• The MCP workspace_list_members tool explains that your role cannot view members, instead of failing generically
• A cached version entry dated in the future is treated as expired, so it can no longer suppress the update check indefinitely
• The path-containment check no longer misreads directory names that begin with two dots, and anchors relative targets on the boundary root

2026-07-27

#0.93.0

#New Features

• Severity grading on review findings - tessl change review now grades every finding nit / minor / major / critical by default; pass --no-grade to turn it off
• Finding attribution - each tessl change review finding names the reviewer skill that produced it
• Explicit review range - new --head flag on tessl change review to review a specific commit range instead of the inferred one
• Publish-access filter - new --has-publish-access flag on tessl workspace list to show only workspaces you can publish to
• Plugin Creator in Tessl Agent - the agent now ships with the Plugin Creator skills for authoring plugins

#Improvements

• Symlink guard for MCP config - the CLI warns and asks for confirmation before writing MCP configuration through a symlink that points outside your project
• Local-execution notice - installing a plugin that ships an MCP command which runs on your machine now tells you at install time
• Global installs respect --agent - the --agent scope is honored when installing context globally
• CLAUDE_CONFIG_DIR support - installing global Claude Code context now honors CLAUDE_CONFIG_DIR
• Clearer permission prompt - the session-scope option in the agent permission prompt has a clearer label
• Clearer MCP review tools - review_run, review_view, and review_fix describe what they do and what to expect
• Keys revoked on logout - tessl logout revokes the device's LLM keys

#Bug Fixes

• Fixed concurrent auto-updates leaving behind a dangling symlink
• Fixed a race when two installs ran at once by staging npm installer downloads
• tessl change verify now fetches the diff base itself in shallow and CI checkouts instead of failing
• Install sync no longer fails when a skill entry is missing from the manifest
• Search results now render when ranking data is unavailable
• Review scores now use the rubric scale reported by the API instead of assuming a fixed scale

#Removals

• tessl context library and its query_library_docs MCP tool have been removed. They were deprecated and have no replacement.

2026-07-17

#0.92.0

#New Features

• Hooks in plugins - Plugins can now declare cross-agent hooks via hooks / nativeHooks in plugin.json. See the plugin configuration docs.

#Improvements

• Signed release verification - CLI installation and self-update now verify against a cryptographically signed release manifest before extraction.

2026-07-15

#0.91.0

#New Features

• Agent-readiness report - run /agent-ready in the tessl agent to scan your repository and surface what to improve.
• tessl launch resume - resume an interrupted or completed cloud launch run.
• Signed release verification - the installer verifies the release signature before installing.

#Improvements

• Agent works without native search tools - uses fd / rg when available and falls back to a basic code implementation when they are not installed.
• Verify cache relocated - now stored at ~/.tessl/verify-cache, so cached results are shared across projects instead of living in each checkout.
• Auto-detect workspace - tessl review run / review fix read the workspace from tessl.json instead of always prompting.
• tessl launch list now shows the workspace and the run's creator.
• tessl eval list consolidates injected context into a single column.
• Credit errors link to top-up - an out-of-credits error now points to the upgrade / top-up flow.
• Explicit, persistent agent setup - tessl init now records selected agents in tessl.json for reproducible installs across machines. Re-run tessl init on an existing project to update the list.
• Clearer permission prompts - prompts are deduped per tool call, bash paths are classified by read/write/delete, external and sensitive-path actions route through the auto classifier, and auto-mode prompt wording is combined.

#Changes

• Separate --agent and --model flags - compound --agent <agent>:<model> tuples are now split into --agent <agent> --model <model>. The old form still works but is deprecated and warns.

#Bug Fixes

• Installer validates the remote release version before installing.
• Generic hook command/args aligned to the exec / shell forms.

2026-07-02

#0.90.0

#Breaking Changes

• tessl review <target> (without run) no longer runs a review - it now shows the review command group's help instead. Use tessl review run <target> (or tessl review run quality <target>).

#New Features

• tessl review run quality / tessl review run security - tessl review run is now a subcommand group: quality (the default, unchanged behavior) scores a skill against a rubric, and security runs a Snyk-powered security scan - replacing the standalone tessl security-review command. tessl review list, view, and retry now work across both kinds from a single unified history, with a --fail-on flag mirroring --threshold for security scans. tessl review run <target> continues to work exactly as before - it's shorthand for tessl review run quality <target>.
• Clone skills from GitLab and Bitbucket sources - Skill installation and inventory scanning now support GitLab and Bitbucket repository sources, not just GitHub.

#Bug Fixes

• Fixed tessl agent ignoring a saved default model from settings.json on startup.
• tessl agent - Fixed session approval matching for external paths when multiple patterns are involved.

2026-06-30

#0.89.0

#New Features

• tessl agent out in Open Beta - It helps you build and optimize your software factory one workflow at a time.
• tessl change commands to aid with automated code reviews
  • tessl change risk - Assess PR/diff risk and decide whether human review is required.
  • tessl change verify - Verify code against configured verifiers.
  • tessl change review - Review the current diff with reviewer skills and emit structured review data.
  • Use tessl agent to set them up!
• tessl launch skill - Run skills using your preferred agent in order to automate parts of your workflow with skills.
• tessl mcp proxy - Connect to MCP servers through an MCP gateway set-up in a workspace’s settings.

#Improvements

• Clearer error when --skill can't apply - tessl install --skill <name> now fails fast with an actionable message when given zero sources or more than one source, instead of silently installing nothing.
• Friendlier model-lock message on eval runs - Selecting a non-default model on a Free plan now produces a single, clear message explaining how to proceed, instead of a raw 403 Forbidden.

#Bug Fixes

• Fixed relative file: sources (e.g. tessl install file:../my-source --global) failing to resolve; they now resolve against your current working directory as expected.
• Fixed a single unsupported entry in a plugin's .mcp.json failing the whole file at publish and install time; unsupported entries are now skipped and the rest of the plugin installs.

2026-06-26

#0.88.2

#Bug Fixes

• npm package - fixup README.md file

2026-06-26

#0.88.1

#Bug Fixes

• Global plugin MCP servers - tessl install --global now writes a plugin's MCP server config to each agent's real user-level config.

2026-06-26

#0.88.0

#New Features

• Plugin MCP Servers - Plugins can now declare MCP servers in a bundled .mcp.json at the package root.
• Publish Through Symlinks - tessl skill publish and tessl plugin publish accept an opt-in --resolve-symlinks flag that copies a symlink target's bytes into the published tarball (within a safe boundary), so authors can reuse files from elsewhere in the repo.

#Improvements

• Workspace Scoping for Reviews - tessl review view and tessl review retry now accept --workspace / -w, and --last resolves the most recent run within that workspace.
• Publish OIDC Warning - When publishing from GitHub Actions without an OIDC token permission, the CLI now warns instead of failing opaquely.

#Bug Fixes

• Fixed tessl skill review --threshold 0 incorrectly emitting a "Skill validation failed" error.

2026-06-23

#0.87.0

#Breaking Changes

• Skill review now requires a Tessl account. Both tessl review and the older tessl skill review now require authentication: sign in with tessl login, or set a TESSL_TOKEN API key in the environment for CI. Reviews run on Tessl's servers, so anonymous review runs are no longer supported.

#Deprecations

• tessl skill review is deprecated - use tessl review instead. tessl review (alias tessl review run) is the replacement: an async, server-side review with companion view, list, retry, and fix commands. The old command still runs for now but prints a deprecation warning on every run and will be removed in a future release. Replace tessl skill review <path> with tessl review <path>.

#New Features

• tessl review is now fully launched as an end-to-end workflow - review runs are first-class in the CLI with run as the default subcommand, and companion commands to view, retry, and fix reviews directly from the terminal.
• Flexible reviewer selection - you can target a specific reviewer with --review-plugin or use your workspace's default reviewer configuration.

#Improvements

• Richer review output for debugging and automation - tessl review view now renders dynamic judges per run, and review JSON includes reviewer/plugin identity metadata.
• Clearer tessl eval view output - run labels are shown and scenarios are sorted for easier reading.
• More informative tessl eval run start output - pre-run context details are surfaced before execution.

#Bug Fixes

• Review command reliability fixes - tessl review list, view, and retry now resolve workspace names correctly, and --json without --workspace now returns a clear error.
• Fixed completed scenario counts in multi-solution eval runs.
• Fixed evaluator failures caused by repeated plugin installation arguments.

2026-06-22

#0.86.0

#New Features

• Customise the eval-scenarios run when publishing - added --eval-scenarios <scenario-search-path> to tessl plugin publish / tessl skill publish. The default is ./evals in the directory of the plugin being published.

#Improvements

• Default eval model is now DeepSeek v4 Flash.
• tessl eval list --context-type - filter eval runs by the type of context injected into scenarios for the "with context" variant.
• Richer tessl eval view - clearer variant labels, an injected-context block, scenario paths, and repo/plugin provenance.
• tessl eval run echoes the scenario search directory and count before starting.
• Commit-fixture uploads are grouped and deduped to remove some cases where redundant context was uploaded.
• Install source policy: git-source allowlist, with denial messages that name the offending source.
• You're prompted to accept pending invites before a personal org is created.

2026-06-17

#0.85.0

#Breaking Changes

• tessl eval run --agent is now single-valued - Previously --agent was repeatable and each value created a separate eval run. It now accepts a single agent; run the command once per agent to compare.
• Removed legacy eval-run context flags - --variant, --context-ref, and --context-pattern have been removed from tessl eval run in favour of the new --context selection (see below).

#New Features

• Skill Inventory - Skill Inventory gives you a living map of every skill in your GitHub org - what exists, where it's used, and where you're duplicating effort. Run the initial import with tessl inventory import and it scans your repos, collects your skills and manifests, and gives you a snapshot of your whole inventory. Run it again any time - each scan is diffed, so you get a living view of what's new, unchanged, and removed over time.
• Context Selection for tessl eval run - A single --context <path> front door now picks the context under test, with --context-commit to source it from a commit and --skill (repeatable) to narrow a local plugin to specific skills.
• Control Baseline by Default - tessl eval run now always runs a control variant to compare against, with --skip-baseline to opt out.
• Secrets injection for plugin evals - tessl eval run now accepts --env-file <path>. Variables in the file are encrypted client-side and delivered into the eval sandbox, where they are available to setup.sh scripts and the agent. See Passing secrets to an evaluation.
• --child Flag for Nested Projects - tessl project create now requires --child to create a project nested inside an existing one, and blocks with a clear explanation when a parent project is detected in an ancestor directory.
• tessl version - Bump a plugin's version locally without publishing, like npm version. Pass patch, minor, or major to increment the current version following semver, or an explicit version (e.g. tessl version 2.1.0) to set it directly. Updates the plugin manifest in place; an optional path argument targets a directory other than the current one.

#Improvements

• Clearer Install Output - tessl install now shows only the dependencies you explicitly requested on the success line; incidental re-installs are summarised as a dim Synced: N count.
• tessl scenario generate workspace resolution - Workspace is now resolved from your linked Tessl project (tessl.json), the --workspace flag, or an interactive prompt — instead of being derived from the tile manifest name, which caused confusing 403 errors when the manifest used the local/ prefix. The --workspace flag is now accepted for tile-based generation.

2026-06-12

#0.84.0

#New Features

• Bootstrap Project Create - tessl project create now works in directories that don't yet have a tessl.json, bootstrapping it as part of the create flow.
• Personal GitHub Account in Skill Inventory Import - tessl skill inventory import can now authenticate with a personal GitHub account, not just org-level credentials.
• Plugin Marketplace Discovery - Tessl now classifies .claude-plugin/marketplace.json as a first-party manifest, picking up plugins from that location.

#Improvements

• Warning for Large-Scale Eval Runs - tessl eval run now prompts a confirmation before launching large eval batches.
• Plan-Entitlement-Gated Model and Agent Selection - model and agent selection respects plan entitlements with clearer messages when a selection isn't available.
• Age-Gated Registry Sources Fallback - registry sources blocked by release-age policy can now resolve to an older fallback source rather than hard-blocking.

#Bug Fixes

• Fixed tessl install to fast-fail on private or missing git repositories instead of silently succeeding with a hidden failure.
• Fixed the judge-results display to show the correct score denominator in both CLI and frontend.

#Windows installer (new distribution channel)

Signed Windows installers are now available for direct download. The Inno installer wraps a signed tessl.exe bundled with TESSL_DISABLE_UPDATE (winget will manage updates):

• https://install.tessl.io/installers-windows/0.84.0/TesslSetup-0.84.0-x64.exe
• https://install.tessl.io/installers-windows/0.84.0/TesslSetup-0.84.0-arm64.exe

The signed package will be installable directly via winget install tessl.tessl as soon as it will be approved by Microsoft

2026-06-09

#0.83.0

#New Features

• **Jujutsu Repo Support - tessl project repair now detects and handles jj-managed repositories alongside Git.
• Finer Eval Run Controls - tessl eval run adds two independent opt-outs: --skip-forced-context-activation and --skip-scoring. Use them on their own or together to control whether the run forces context activation and whether the rubric scorer fires.

#Improvements

• Solver and Scorer Flag Deprecation - The --solver and --scorer flags on tessl eval run are now hidden but still functional, with a one-line deprecation warning; prefer the new --skip-* flags.

#Bug Fixes

• **Removed a vendor-native plugin-manifest ingestion path that could pick up stale manifests outside the expected resolution flow.

2026-06-05

#0.82.0

#Improvements

• Install Policy Enforcement - tessl install now honors org-configured release-age and commit-age policies. For registry sources, the CLI automatically falls back to an older version that satisfies the policy (with a clear reason in the install output) or hard-blocks if no version qualifies. For git sources, it resolves to the most recent commit old enough to satisfy the minimum commit-age policy instead of installing HEAD.
• Policy-Aware tessl outdated - Versions blocked by a release-age policy are now annotated as [release age policy] in the update table, and git sources show the latest age-compliant commit as the recommended update so you can see what will (and won't) be installed before running tessl update.

2026-06-03

#0.81.3

#Improvements

• Eval Failure Reasons - eval run and eval view now display the reason when a run fails (e.g. quota exceeded), instead of a bare failure marker
• Eval Default Model Notice - A one-time notice informs you that the default eval model is now GLM 5.1, with guidance on selecting a different model
• Resilient Eval Polling - eval run now rides out transient server errors and network blips while polling, instead of reporting a completed run as failed
• Clearer Validation Errors - Manifest schema errors are now rendered in a readable format instead of a raw JSON dump
• MCP Tool Updates - The new_tile MCP tool is renamed to new_plugin, and the outdated tool now accepts a projectDir parameter to check a specific directory

#Bug Fixes

• Fixed plugin pack failing with a missing tile.json error for plugins created with plugin new; migrated plugins also no longer pack from the stale legacy manifest
• Fixed eval run evaluating the stale legacy tile.json instead of .tessl-plugin/plugin.json when both manifests are present
• Publishing a migrated plugin now errors if tile.json declares private: true but plugin.json omits it, preventing a private plugin from accidentally going public
• Fixed broken documentation links shown after plugin new and skill new
• Removed misleading advice pointing at tile.json from the "no content" validation error

2026-05-30

#0.81.2

#Bug Fixes

• scenario generate: Clearer Error When No Manifest Found — Running tessl scenario generate in a directory with no manifest previously crashed with a confusing ENOENT: tile.json error. It now shows a clear message pointing you to add a .tessl-plugin/plugin.json.

2026-05-30

#0.81.1

#Bug Fixes

• Fixed skill publish incorrectly triggering the import wizard when plugin.json already exists
• Fixed skill publish --bump not writing the bumped version to plugin.json, causing the old version to be published
• Fixed skill publish pre-publish version checks reading from tile.json instead of plugin.json

2026-05-29

#0.81.0

#New Features

• Tiles are now Plugins - This release completes the switch from Tiles to Plugins. The CLI now uses "plugin" across every command, help listing, flag description, and message. For more information on how to migrate from Tiles to Plugins, there is a guide in the documentation.
• Full moderation status in plugin info - plugin info now shows the complete moderation state for a published version (pass, pending, skipped, fail, or error).

#Improvements

• ls shortcut - tessl ls now works as a shortcut for tessl list, mirroring the existing i → install shortcut.
• Workspace-scoped uploads - Eval-related uploads (scenario tarballs, plugin archives, repo archives) are now scoped to the active workspace, so teammates in the same workspace can access uploads created by other members.

#Bug Fixes

• Eval run links now point directly to workspace-scoped URLs, skipping the redirect members previously hit when clicking through.
• scenario generate now works in plugin-only directories (a .tessl-plugin/plugin.json with no tile.json), which previously failed with a "tile.json not found" error.
• The CLI no longer fails to start when its log file can't be written (e.g. in sandboxed or read-only environments) — file logging is now best-effort.

2026-05-21

#0.80.0

#New Features

• Eval Enablement Checks - tessl scenario generate, tessl tile publish, and tessl skill publish now pre-flight check whether the target workspace has evals enabled. Scenario generation fails fast with a clear message; publish still succeeds but cleanly skips the eval upload with a single info line.

#Improvements

• Cleaner Publish Output - tessl tile publish no longer prints validation warnings twice (once from pre-publish validate, once from pack).
• Clearer Forbidden Errors - When creating a project fails because the workspace doesn't exist or you lack permission, the CLI now shows an actionable message instead of a generic 403 Forbidden.
• Skill Install Paths - Skills are no longer written under .gemini/skills; installation routes through .agents/skills, and any legacy tessl:* / tessl__* entries left in .gemini/skills are cleaned up automatically.

#Bug Fixes

• tessl tile new now preserves the rule description you provide instead of overwriting it with a TODO placeholder.
• Generated RULES.md now ends with a trailing newline.

2026-05-12

#0.79.1

#Improvements

• Clearer Guidance for Missing Project Links - When a linked Tessl project can't be found or isn't accessible, the CLI now surfaces actionable next steps with specific tessl project create and tessl project repair commands instead of a raw 404. The tessl eval run help text now also explains that eval runs are saved to a Tessl project.

2026-05-11

#0.79.0

#New Features

• OpenHands agent support (alpha) — tessl init, tessl install, and tessl run now recognise OpenHands. Auto-detection follows the standard project → home → PATH ladder; OpenHands is excluded from the interactive picker while in alpha but can be selected explicitly via --agents openhands.
• tessl search --global — install search results into ~/.tessl/ without requiring a project to be initialised. Short form g available.
• tessl eval run --list-agents — print the supported <agent>:<model> catalogue (with the default marked) without running an eval. Supports --json for scripting.
• tessl project - Tessl now supports workspace-backed projects in the CLI, giving your local work a persistent project identity for evals and other project-based workflows. Use tessl project create to create a project, tessl project link to connect an existing one, and tessl project repair to fix broken project links.
• tessl eval run --scorer-agent — override the agent and model used for the score step (as agent:model), while the solver agent stays unchanged. tessl eval view shows the override under a Scored by: line.
• tessl eval run --quality-check — opt-in flag that asks the backend's LLM judge to filter out misleading scenarios (low feasibility, rubric leakage, low value) before solve and score. Excluded scenarios show up under a red ✗ Excluded by quality check line in tessl eval view. tessl tile publish --with-scenario-quality-check enables the same filter for the eval that runs as part of publishing.

#Improvements

• Simpler tessl skill import / skill new — no longer prompts for workspace or visibility; defaults to local workspace and private visibility. Pass flags explicitly to override.
• Scenario config (scenario.json) — declarative working-directory setup for evals. Declare fixtures to install (commit or directory), include paths to copy in verbatim, and setup scripts to run after install. Conventional defaults (resources/ auto-included, setup.sh auto-run) mean most scenarios need no scenario.json at all. tessl scenario generate now produces a scenario.json for every generated scenario, plus any supporting input state (e.g. resources/) when the source commit needs it. Setup scripts are off by default per workspace — contact Tessl to enable.
• Cleaner install error messages — local paths missing the file: prefix now surface a clear validation error instead of a generic parse failure.
• Softer security-review copy — surfaces showing a per-skill security status now display "No security review available." instead of "Security review has not been run yet."
• tessl skill review includes referenced skill content — skill reviews now include all referenced content and scripts for logged-in users.

#Bug Fixes

• Install exits non-zero on failure — tessl install and tessl install --global now reliably propagate tile install failures as command failures (non-zero exit code).
• tessl scenario generate <relative-path> — relative paths like tiles/foo are now correctly routed to the tile flow when the directory exists locally, instead of misrouting to the repo flow and erroring on missing -commits/-prs.
• tessl tile lint — @decorator tokens in markdown headings and YAML frontmatter are no longer treated as hard-link references and no longer produce false-positive lint warnings.
• MCP install bootstrap message — the "Initialising new tessl project at …" message no longer prints during bulk sync installs that don't actually write tessl.json.
• Experimental agent detection — explicit project-level usage of experimental agents (e.g. OpenClaw workspace markers) is now respected ahead of the experimental gate, and the default install flow no longer force-enables experimental agents.

2026-04-29

#0.78.0

#New Features

• Activation Solver for Evals - tessl eval run --solver=activation measures which skills activate for each scenario, and tessl eval view renders the results in a per-scenario activation table
• Set Project Name on Init - tessl init accepts a new --name flag to set the project name in tessl.json at creation time
• Target Org on Workspace Create - tessl workspace create now accepts a --org-id flag to associate a workspace with a specific organization
• List Organizations - New tessl org list command displays the organizations you belong to
• Scenario Quality Checks for Evals - tessl eval run --quality-check will filter out scenarios that would give misleading results, based on a quality assessment that looks for things like “criteria leaking into the task” or “task duplicating information from the skill it is meant to exercise”

#Improvements

• Clearer Eval Run Status - Pending and in-progress eval and generation runs render as yellow "In Progress" with elapsed time, instead of an ambiguous "Pending"
• Better Invalid Manifest Errors - tessl eval run against a directory with a malformed tile.json now reports the manifest error directly with a hint to fix it, instead of falling through to the scenarios flow
• Removed Eval Compare - tessl eval compare has been removed. Use tessl eval view and tessl eval list instead
• Global Install from Search - tessl search now supports --global so selecting a result installs it to ~/.tessl/ instead of the current project

#Bug Fixes

• Fixed tile rule creation to safely handle prompts containing / so they no longer create unintended nested paths in the generated rule files

2026-04-22

#0.77.0

#New Features

• Experimental Windows Support - The Tessl CLI is now available on Windows as an experimental release. Install it with PowerShell:

  irm https://install.tessl.io/scripts/latest.ps1 | iex

• Eval Setup Scripts - Eval scenarios support a new setup field in scenario.json — an ordered array of script paths that run after fixtures and includes are installed, before the solver agent starts.

#Improvements

• Cleaner Top-Level Help - Simplified top-level CLI help.

2026-04-14

#0.76.0

#New Features

• Eval Progress Bars — eval run and eval view now show real-time per-scenario progress bars with colour-coded status (complete, active, queued)
• Directory Fixtures in Evals — Scenarios can now declare include paths for directory-based fixtures, alongside existing commit-based fixtures.
• "latest" Dependency Version — Use "latest" as a dependency version in tessl.json to always resolve the most recent published version
• Uninstall Individual Skills — tessl uninstall --skill <name> removes specific skills from a dependency instead of removing the entire tile
• Verifiers in Tile Packing — The verifiers/ directory is now included when packing tiles (when evals are enabled)
• Failed Security Review Visibility — Security reviews that failed to run are now shown as "failed" instead of appearing as "pending"

#Improvements

• Project Name from Directory — tessl init now derives the project name from the directory basename instead of using a generic default
• Archive Without Version — tessl archive now works without a version suffix, allowing you to archive all versions of a tile at once
• Eval Agent/Model Display — eval view now shows the agent and model for all eval run types
• Workspace Commands — -user flag renamed to -username , -user still works with a deprecation warning
• Help Text Corrections — Improved descriptions for install --skip-init, tile pack, skill import, and config set

#Bug Fixes

• Fixed eval scenario path resolution when base directory is "."
• Fixed eval git diff ignoring user's custom diff tool configuration
• Fixed object metadata serialization in eval debug output

#Breaking Changes

• Removed --project-dependencies — The --project-dependencies flag has been removed from tessl install and tessl init. Dependency scanning and automatic tile matching are no longer part of the install flow. Use tessl search to find relevant tiles instead.

2026-03-30

#0.75.0

#New Features

• API Key Management — New api-key command group with create, list, and delete subcommands for managing workspace-scoped API keys. Supports both interactive and non-interactive (scripting) modes
• Auto-Select Single Workspace — When you're a member of only one workspace, the CLI now skips the workspace selection prompt and selects it automatically

#Breaking Changes

• auth token Removed — The auth token command has been removed. Use the new api-key create, api-key list, and api-key delete commands instead for workspace-scoped API key management

2026-03-26

#0.74.0

#New Features

• Machine-Readable Search: tessl search --json flag outputs results in JSON format for scripting and automation
• Skill Review Threshold: tessl skill review now supports a --threshold option: when the score is below this threshold, the command will fail - helpful for CI use cases
• Scores in Search & Info: Search results and tessl tile info now display quality scores and security review details

#Improvements

• Piped Token Output: tessl auth token now suppresses the spinner and hints when output is piped, making it easier to use in scripts
• Better Error Display: Failed operations now render a red cross indicator for clearer error visibility
• Search Column Alignment: Interactive search results now have properly aligned columns
• Remove Steering to Use query_library_docs MCP tool: It was getting in the way of skill activation from some users. You can manually opt-in to install the rules yourself with tessl install tessl/cli-setup

#Bug Fixes

• Fixed skill review API calls failing due to stale authentication tokens
• Fixed eval commands resolving git repo root incorrectly when run from subdirectories

2026-03-16

#0.73.0

#New Features

• Security Reviews on Install - tessl install and tessl update now show security review results before installation. CRITICAL/RISKY findings require confirmation to proceed, ADVISORY findings show an advisory after install, and unscanned items display a warning. Use -dangerously-ignore-security to bypass prompts in CI/scripted contexts
• Commit-Pinned Installs - tessl install now accepts github:<repo>@<sha> and full GitHub commit URLs, allowing you to install a skill at a specific commit. A post-install hint directs to tessl update for future updates
• Publish --bump Flag - tessl tile publish and tessl skill publish now accept --bump patch|minor|major to auto-increment the version when the current version already exists in the registry
• .agents/skills Directory Support - Install skills in .agents/skills alongside other requested agents since that’s the folder many tools are aligning with

#Improvements

• Grouped Search Results - tessl search now groups results by type (Tiles / Skills) with section headers when both types are present

#Bug Fixes

• Fixed tessl skill new showing interactive prompts even when all required flags (-name, -description, -workspace) are provided
• Fixed MCP server processes becoming orphaned when the parent process disconnects (stdin closes)

2026-03-12

#0.72.0

#New Features

• Pre-Publish Validation & Dry Run - Tile and skill publish commands now validate permissions, workspace, and version availability before publishing, with clear pass/fail output. Use --dry-run to check without actually publishing.
• Security Scores in Search - Search results now display security assessment levels (Passed, Caution, Risky, Critical) alongside tile scores

#Improvements

• Smarter Eval Scenario Generation - Existing scenarios are now considered when generating new ones, avoiding duplicates
• Large File Exclusion in Eval Archives - Files over 10MB are automatically excluded from repo archives during eval runs
• Eval Commit Ref Validation - Commit refs are validated before archive creation, with clear error messages for missing refs

2026-03-11

#0.71.0

#New Features

• Support for Copilot in VSCode — Skills, rules, and our MCP server can be automatically set-up for vscode
• Eval Run Labels — New --label / -l flag on tessl eval run lets you attach a short description to a run. Labels appear in tessl eval list and the Tessl web UI, making it easier to identify what you were testing across multiple runs.
• Eval Variant Selection — New --variant flag on tessl eval run lets you run only a specific variant. Pass --variant without-context to run the baseline only (useful for iterating on scenario quality without affecting the with-context comparison), or --variant with-context to skip the baseline.
• Local Repo Eval Runs — Project evals can now run against local git repos without SCM integration; the CLI archives and uploads the repo automatically
• Smarter Scenario Generation — tessl eval generate now avoids duplicating existing scenarios when generating new ones; --count specifies how many new scenarios to add

#Bug Fixes

• Allow installation of skills from unknown dot folders in GitHub repositories (e.g., .agent-skills/)

2026-03-10

#0.70.1

#Improvements

• Non-Interactive Mode for Evals & Scenarios — tessl eval run and tessl scenario generate now print the run/generation ID and URL immediately when running non-interactively, instead of polling for completion
• Clearer Help Text — tessl eval run and tessl scenario generate now show which flags are mode-specific, and invalid flag combinations are rejected with helpful messages

#Bug Fixes

• Fixed install --watch-local exiting prematurely — the process now stays alive until Ctrl+C and updated output to show relative paths rather than absolute ones
• Fixed installation from file: source to exclude .git directories - It now installs the same files that would be captured by tessl tile pack or tessl tile publish

2026-03-06

#0.70.0

#New Features

• Local Tile Watching — New --watch-local flag on tessl install watches local file-source tiles for changes and automatically reinstalls tiles, streamlining the tile development workflow
• Search Score Display — tessl search results now show aggregate quality and impact scores with color-coded badges
• Scenario Generation from PRs — New --prs flag on tessl scenario generate accepts PR numbers as an alternative to --commits
• Multiple Scenario Downloads — tessl scenario download now accepts multiple space-separated IDs (e.g., tessl scenario download id1 id2)
• Eval Run Repeats — New --runs / -n flag on tessl eval run to repeat each agent configuration multiple times
• Eval Polling for Tile Evals — Tile evals now poll for completion instead of returning immediately, with Ctrl+C to detach gracefully
• Eval JSON Output for Tile Evals — --json flag on tessl eval run now works for tile evals, outputting run ID, tile name, and scenario count
• Eval Agent Flag — New --agent flag (format: agent:model) replaces --model for tile evals, aligning with project eval conventions
• Eval Cost Metadata — tessl eval view now displays cost, runtime, turns, and token usage below each solution
• Eval Agent/Model Display — tessl eval view stats now show the agent and model used for each solution
• --last for Eval Debug & Retry — tessl eval debug and tessl eval retry now support --last for quick access to the most recent eval run
• Repository Metadata in tile.json — Tiles now support an optional repository URL field in their manifest

#Improvements

• Tile Cache Protection — Tile authoring commands now warn when operating on paths inside .tessl/tiles/, preventing accidental edits that would be overwritten on the next install
• Cleaner Help Text — Removed meaningless --no-{flag} negated boolean flags from all commands, improved flag capitalization, and reordered commands for logical grouping
• Consistent Scenario Terminology — "Generation run" renamed to "generation" throughout the CLI for simpler, more natural language
• Scenario Generate Table Output — tessl scenario generate results now render as a formatted table with Scenario ID and Commit ID columns
• Smarter Binary Install — Skips re-downloading binaries that already exist and pass validation; corrupt binaries are automatically cleaned up

#Fixes

• Fixed CLI crashes when output streams disconnect unexpectedly (EPIPE handling)
• Fixed unpublish help text to reflect the correct 2-day timeframe (was showing 2 hours)

2026-03-02

#0.69.0

#New Features

• Scenario Generation Commands - New tessl scenario command suite for generating and managing test scenarios. Generate scenarios from tiles or repository commits (tessl scenario generate), download them to disk (tessl scenario download), and track generation runs (tessl scenario list, tessl scenario view)
• Global Tile Installation - Install tiles globally with tessl install --global to make documentation and skills available across all your projects.
• Eval Variant Comparison - tessl eval run now supports running evaluations with multiple agent variants and context patterns, with automatic detection of tile vs. project scenario sources

#Improvements

• Project Directory Hints - tessl init and tessl install now display a helpful hint when the target project directory differs from your current working directory

#Bug Fixes

• Fixed a crash when piping CLI output to another command (e.g., tessl list | head) by gracefully handling broken pipe signals
• Fixed markdown link validation incorrectly parsing non-markdown files during tile linting

2026-02-25

#0.68.0

#New Features

• GitHub Copilot CLI Support — Tessl now recognizes GitHub Copilot CLI as an agent, with automatic detection, rules wiring via AGENTS.md, skill installation, and MCP configuration
• Search Type Filtering — tessl search now accepts a --type flag to filter results by content type (skills, docs, or rules)
• Skill Review Auto-Accept — tessl skill review --optimize now supports a --yes / -y flag to skip the confirmation prompt and auto-apply improvements

#Improvements

• Automatic Binary Cleanup — Old CLI binary versions are now automatically cleaned up after updates, keeping only the current version and one rollback
• Short SHAs for Git Tiles — tessl list and tessl doctor now display short SHAs for git-sourced tiles instead of full commit hashes
• Command Detection Timeout — Command lookups now have a 1-second timeout to prevent indefinite hangs during agent detection

#Breaking Changes

• tessl skill search Removed — Use tessl search --type skills instead

#Bug Fixes

• Fixed tessl skill review --optimize failing with a raw 401 error when not authenticated — now shows a clear login prompt
• Fixed unpublish help text to reflect the correct 2-day window (was incorrectly showing 2 hours)

2026-02-23

#0.67.0

#New Features

• Skill Optimization - tessl skill review --optimize now applies feedback from the review to generate an updated SKILL.md file
• Evaluation Type Support - tessl eval list now displays all evaluation types (tile, repo, skill) with a new "Type" column and --type flag to filter by evaluation type
• Filter Evaluations by Status - tessl eval list now supports a --status flag to filter runs by status (pending, in_progress, completed, failed)

#Bug Fixes

• Fixed tessl skill review returning exit code 0 on validation failure — it now correctly returns a non-zero exit code, enabling proper CI/CD integration
• Fixed tessl api --input - sending incorrect Content-Type when piping JSON via stdin

2026-02-22

#0.66.0

#New Features

• outdated command — Check installed tiles for newer versions, showing current, safe update, and latest version info
• update command — Update tiles to newer versions. Use --yes for all, and --force to include breaking updates

2026-02-18

#0.65.0

#Bug Fixes

• Windows Compatibility for Generated Files - Fixes generated file and directory names containing colons (:) which are invalid in Windows file paths. Tile files committed to git repositories can now be cloned on Windows without errors. Existing installations are migrated automatically

2026-02-17

#0.64.0

#New Features

• Local Tile Evaluation - Run evals against your tiles directly from the CLI with tessl eval run and get a fast feedback loop for improving your tiles
• Eval Results in Terminal - View eval results with per-scenario breakdowns, variant comparisons, and scoring summaries using tessl eval view
• Eval Run History - List recent eval runs filtered by workspace, tile, or author with tessl eval list
• Eval Retry - Re-run failed evals with tessl eval retry

2026-02-13

#0.63.4

#Improvements

• Internal - Stability improvements

2026-02-13

#0.63.3

#New Features

• Local Path Support for Tile Info - tessl tile info now accepts a local directory path as an argument and defaults to the current directory when no argument is provided, making it consistent with other tile commands like publish and lint

#Improvements

• Better Publish Messages - After publishing tiles or skills, the CLI now displays the published resource URL and a note about moderation timelines

#Bug Fixes

• Fixed MCP configuration failures when using npx tessl by ensuring the tessl binary is accessible via the PATH during init

2026-02-11

#0.63.2

#Bug Fixes

• Fixed circular reference issues when CLAUDE.md and AGENTS.md are symlinked

2026-02-11

#0.63.0

#Improvements

• The CLI is now distributed as a native binary instead of a bundled Node.js package

2026-02-10

#0.62.2

#New Features

• Optional Authentication - You can now use tessl init, tessl install, and tessl search without logging in. Public tiles are accessible to everyone, while private tiles will show helpful hints to log in when needed

#Bug Fixes

• Fixed --skill flag not working correctly when installing a single skill from a local tile with multiple skills

2026-02-09

#0.62.1

#Improvements

• Internal - Stability improvements

2026-02-06

#0.62.0

#New Features

• Unified Search — tessl search now returns both tiles and skills in a single view, with visual badges showing content types ([docs], [skills], [rules]) and skill provenance
• Tile Evaluations — New tessl eval command group for evaluating tiles:
  • tessl eval run [path] — Run evaluations on local tiles
    • --force flag to re-run all evaluations including previously solved cases
  • tessl eval list — View recent eval runs with status
  • tessl eval view-results <id> — View detailed results with color-coded scores and assessment tables
    • --last flag to quickly view most recent eval results
  • Git metadata (branch, commit, dirty state) automatically attached to eval runs
• Tile Info — New tessl tile info <name> command to view tile metadata, verification status, and moderation status
• Skills in MCP Context — The query_library_docs MCP tool now includes installed skills when building knowledge context

#Improvements

• Redesigned Help for tessl -h — Commands are now grouped by category with improved formatting and readability
• Publish Workspace Context — Publishing now shows the target workspace name before uploading

#Bug Fixes

• Fixed --skill flag being ignored when installing skills from registry tiles
• Fixed re-installing tiles not updating the manifest with new skill selections
• Fixed tile lint incorrectly flagging files as orphaned when linked via @path.md includes

2026-02-02

#0.61.4

#Improvements

• Install --yes Flag Behavior - The --yes flag now auto-selects all skills when installing from GitHub, even in interactive mode. Previously it only worked in non-interactive environments.
• Non-Interactive tile new - The tessl tile new command can now run fully non-interactively when all required flags are provided, making it easier for scripts and agents to use.

#Fixes

• GitHub Skill Installation Bugfixes - Address various edge cases that prevented some skills being installed from GitHub
• GitHub Skill Installation Respects Pinned Version - Fixed an issue that meant we always installed a skill from a repository's trunk rather than the commit sha in tessl.json when doing tessl install

2026-01-31

#0.61.3

#Improvements

• Faster git installs - Git dependencies are no longer re-downloaded on every install when already at the correct version

2026-01-30

#0.61.2

#Improvements

• Windows Platform Detection - The CLI now detects Windows and provides a clear error message directing users to use WSL instead
• Lenient GitHub Skill Installation - Skills from GitHub with missing referenced files or minor schema issues now install with warnings instead of failing completely
• Better Warning Display - Installation warnings are now grouped by tile name with count summaries (e.g., "3 content warnings in github:user/repo")
• Verbose Install Output - Added --verbose (-v) flag to tessl install and tessl skill search commands to display detailed warning messages during installation

#Bug Fixes

• Fixed skill validation warnings showing absolute temp directory paths instead of relative paths
• Fixed installation failures when GitHub skills have invalid optional frontmatter fields (e.g., allowed-tools as array instead of string)

2026-01-29

#0.61.1

#New Features

• Skill Commands - Added a complete set of dedicated skill management commands: tessl skill new, tessl skill import, tessl skill lint, tessl skill publish, tessl skill review, and tessl skill search. Skills are now first-class citizens with their own command group for easier discovery and management.
• GitHub Installation Support - Install tiles and skills directly from GitHub repositories using tessl install github:user/repo or full URLs. Supports branch selection, subdirectories, and interactive skill selection. Git sources are pinned to commit SHAs for version stability.
• Skill Search - Discover available skills across the Tessl ecosystem with the new interactive tessl skill search command. Search returns both external skills from GitHub and tiles containing skills, with interactive selection and installation.
• Skill Review - Evaluate skill quality with tessl skill review, which validates SKILL.md files and provides LLM-based scoring across multiple dimensions with actionable suggestions for improvement. Supports both local paths and GitHub URLs.
• Non-Interactive Mode - Commands now auto-detect TTY availability and work in CI/CD pipelines and with AI agents. When required flags are missing in non-interactive environments, commands provide clear error messages pointing to available options.
• Selective Skill Installation - Choose specific skills when installing tiles using the new include.skills manifest field in tessl.json. Reduces clutter in agent directories by only activating the skills you need.
• Codex Support - Full support for Codex as a core agent alongside Claude Code, Cursor, and Gemini. Includes project detection, MCP configuration, steering, and skills synchronization.
• Gemini Production Ready - Gemini agent is no longer experimental and fully supported in production workflows.

#Improvements

• Simplified Skill Naming - Skill directories now use cleaner names (tessl:<skill> instead of tessl:<workspace>:<tile>:<skill>) and are symlinked rather than copied for better performance and consistency.
• Automatic Eval Publishing - Evals are now published by default when running tessl tile publish or tessl skill publish. Use -skip-evals to opt out. Published evals automatically trigger evaluation runs.
• Better Skill Discovery - Improved skill detection uses a three-step process (root check, priority directories, full scan) with optimizations that reduce scan time and exclude build/cache directories.
• Streamlined Publishing - The tessl skill publish command no longer asks redundant "will you share this?" prompts since publishing inherently means sharing.
• Improved Help Text - Updated tessl init --agent to list all supported agents (claude-code, codex, cursor, gemini) and clarified that tessl install accepts both registry tiles and GitHub URLs. Added tessl i as a shortcut alias.
• Relative Paths in Manifest - When creating tiles with --install, the CLI now stores relative paths in tessl.json instead of absolute paths, making projects portable across machines and directories.
• Smarter CLI Setup - The tessl/cli-setup tile is now only installed when your tiles have documentation, reducing overhead for steering-only projects.
• Flexible Tile Content - Tiles can now contain any file type, not just markdown files, allowing for configuration files, code examples, and other documentation formats.

#Bug Fixes

• Fixed skill review score display showing /4 instead of the correct /3 scale
• Fixed status command handling of non-semver versions for git-based and file-based tiles
• Fixed orphaned file detection for tiles with root-level skills - README.md and other root files are no longer incorrectly flagged as orphaned
• Fixed skill publish command to accept SKILL.md file paths directly instead of only directory paths
• Fixed file source tiles being incorrectly deleted as orphaned on repeated installs when dependency key differed from tile name
• Fixed skill frontmatter validation warnings during installation from external sources like GitHub
• Fixed tessl skill new output messages to reference correct tile directory paths
• Fixed skill selection in GitHub repos to require explicit choices, preventing accidental installation of all skills from large repositories

2026-01-27

#0.60.0

#New Features

• Project Mode (Managed vs Vendored) - Choose how tile content is managed in your project:
  • Managed mode - Tile contents are gitignored like node_modules (default for existing projects)
  • Vendored mode - Tile contents are committed to your repository (default for new projects)
  • Switch modes anytime by setting "mode": "managed" or "mode": "vendored" in tessl.json

#Improvements

• Simplified Local Tile Dependencies - Local tiles (using file: or link: sources) no longer require a version field in tessl.json. The CLI automatically reads versions from installed tile.json files. Existing manifests with both version and source are automatically migrated to the new format.

2026-01-26

#0.59.0

#New Features

• .tileignore Support - Exclude files from tile validation and packaging using gitignore-style patterns.

2026-01-23

#0.58.1

#Improvements

• Tile Commands Auto-Detect - tessl tile pack, tessl tile lint, and tessl tile publish can now be run directly from your tile directory without explicitly specifying the path

#Bug Fixes

• Squashed some bugs that were hiding in the codebase. They know what they did.

2026-01-22

#0.58.0

#New Features

• Context Cost Analysis (Experimental) - The tessl doctor and tessl tile lint commands now display estimated token costs for tile content loaded into agent context windows. This helps you understand which tiles contribute most to context usage and identify oversized content before publishing.
• HTTP Proxy Support - The CLI now works in proxied network environments by automatically detecting and using HTTP/HTTPS proxy settings from environment variables (http_proxy, https_proxy, HTTP_PROXY, HTTPS_PROXY). The CLI respects no_proxy/NO_PROXY bypass rules.

#Improvements

• Enhanced Python Dependency Scanning - Improved Python dependency detection with support for extras notation (e.g., openai[aiohttp]>=1.52.2) across requirements.txt, pyproject.toml, and uv tree output. The scanner now also automatically falls back to pip3 when pip is not available, improving compatibility with macOS and Linux distributions.

#Bug Fixes

• Fixed Go package PURL validation to require the v prefix in version strings (e.g., v1.9.3 instead of 1.9.3), ensuring pkg.go.dev documentation links work correctly.
• Fixed false "broken link" warnings during tessl tile lint when markdown files contain URI schemes like mailto:, tel:, ftp:, or data:.

2025-12-19

#0.57.0

#New Features

• Monorepo Subfolder Support - The tessl install --project-dependencies command now scans dependencies from your current working directory, enabling you to run it from monorepo subfolders in an initialized project (e.g., tessl init && cd backend && tessl install --project-dependencies)

#Improvements

• Smarter Documentation Tree - The context tool now only includes files that are transitively linked from a tile's documentation entrypoint, filtering out orphaned files and making documentation discovery more focused

#Bug Fixes

• Fixed markdown link validation to properly handle anchors (e.g., ./guide.md#section). The validator now validates both file existence and anchor validity, with clearer error messages for broken links

2025-12-16

#0.56.2

#Improvements

• Internal - Stability improvements

2025-12-12

#0.56.1

#Improvements

• Observability Enhancements - Improved telemetry infrastructure for better diagnostics and system monitoring

2025-12-11

#0.56.0

#Breaking Changes

• Node.js 18 No Longer Supported - The CLI now requires Node.js 20 or later. Node.js 18 reached end-of-life in April 2025 and is no longer supported. The CLI will now fail with a clear error message if you attempt to run it on Node.js 18.

#Improvements

• Only Auto-Init Outside Existing Projects - The install and search commands no longer automatically initialize a project when run in a subdirectory of a Tessl project directory. This prevents accidentally creating unwanted projects when running commands after changing directories. To initialize a project, explicitly run tessl init first.

#Bug Fixes

• Fixed an issue where the init command's failed to identify and install relevant tiles.

2025-12-10

#0.55.1

#Improvements

• Faster Auto-Updates - The CLI now checks for updates every 3 hours (down from 24 hours), ensuring you get critical fixes and improvements more quickly. You can still adjust the interval with TESSL_AUTO_UPDATE_INTERVAL_MINUTES or disable it by setting it to 0.

2025-12-09

#0.55.0

#New Features

• Updated Workspace Roles - Workspace member roles have been updated to: member, publisher, manager, and owner (the viewer role has been removed)

#Bug Fixes

• Fixed authentication issue where API token authentication (TESSL_TOKEN) would fail after token refresh attempts, causing authenticated API calls to fail

2025-12-09

#0.54.0

#New Features

• Tile Manifest Validation - The describes field in tile manifests is now validated to ensure it contains a valid PURL (Package URL) format. Invalid PURLs or unsupported package types will show helpful warnings during tessl tile lint, tessl tile pack, and tessl tile publish operations

#Improvements

• Project Directory Detection - Improved project directory discovery logic allows the CLI to better locate your tessl.json even when running commands from subdirectories within your project

#Bug Fixes

• Fixed missing newline after tessl list output for better terminal formatting

2025-12-03

#0.53.0

#New Features

• Go Modules Support - Dependency scanner now detects Go projects and extracts dependencies from go.mod files
• Auto-Init on Search - When selecting a tile from search results, the CLI now automatically initializes the project if needed (creates tessl.json)
• Dynamic Terminal Sizing - Interactive prompts now adapt to your terminal height for better readability
• Escape Key Cancellation - Press Escape to cancel select/checkbox prompts

#Improvements

• Smarter Context Tool - When no documentation is found, the tool now suggests running a search to find relevant tiles
• CI Environment Detection - Auto-update and version checks are now automatically disabled in CI environments

#Bug Fixes

• Fixed tile installation failures when system temp directory is on a different filesystem (handles cross-device moves)

2025-11-28

#0.52.2

No changes in this release

2025-11-28

#0.52.1

#Improvements

• Homebrew installation now available for stable releases
• MCP server stability improvements

2025-11-28

#0.52.0

#New Features

• Added tessl tile lint command to validate tiles before publishing

#Bug Fixes

• Tile upgrades now properly clean up old files when installing new versions
• Fixed markdown link validation to skip code blocks in tile documentation

2025-11-26

#0.51.0

• MCP Tool Improvements
  • Renamed get_library_context tool to query_library_docs and updated descriptions for better clarity
  • Updated cli-setup steering tile version to reflect tool rename and improve adoption
• Tile Management
  • Added tessl list command to display installed tiles
• CLI UX
  • Improved CLI message spacing and copy for better readability

2025-11-19

#0.50.6

• Added the ability to archive and unarchive workspaces
  • Archiving a workspace allows you to non-destructively remove workspaces you no longer wish to use or see. Archiving a workspace also archives all tiles in that workspace. This means that users can still install those tiles if defined in their manifest, but will not otherwise see them in search etc.
  • Unarchiving, as you’d expect, shows a workspace once again. However, it does not unarchive tiles within the workspace, to avoid potentially unarchiving content that was previously archived before the workspace was.
• Fixed a bug when authenticating in multiple machines that would result in our MCP tool get_library_context only working on the last machine you authenticated.

2025-11-18

#0.50.5

We added a new tessl tile pack command which validates and builds a tarball for a tile. It can be useful to validate things before doing tessl tile publish.

Also fixed a bug where the CLI would fail to run when who signed up without a first or last name.

2025-11-17

#0.50.4

Fix tessl tile publish requiring an absolute path

2025-11-14

#0.50.3

This release marks a significant shift in focus. We've rebuilt the CLI from the ground up as an Agent Enablement Platform, pausing work on the Tessl Framework to concentrate on our registry functionality and agent integration.

We really appreciate all the feedback we have received so far. Please, keep it coming.

#What's Changed

#Getting Started is Now Simpler

When you run tessl init in your project, it now automatically detects your dependencies and pulls relevant tiles from the registry. For teams already using tessl, new developers only need to install the CLI and grant MCP server permissions—everything else happens automatically when the MCP starts up.

#Smarter Agent Integration

Instead of agents searching through generated documentation, they now use our new get_library_context MCP tool to query tiles directly. This requires logging in to tessl, but makes context retrieval much more efficient. The MCP tools also include login (for in-session authentication) and status (for checking project state).

Note: KNOWLEDGE.md is no longer generated since agents now pull context on-demand.

#Command Reorganization

We've streamlined the command structure. Changes to existing commands:

• tessl registry install|search → tessl install|search
• tessl registry sync → tessl install --project-dependencies
• tessl registry publish|unpublish|archive → tessl tile publish|unpublish|archive
• tessl version → tessl --version

#Release Channels & Auto-Updates

You can now choose between latest (default, stable) and beta (early access) channels. The CLI checks for updates hourly and auto-updates transparently (max once per 24 hours). Configure the interval with $TESSL_AUTO_UPDATE_INTERVAL_MINUTES or set it to 0 to disable.

#New Commands

• tessl doctor - diagnose authentication, manifest status, detected agents, and dependencies (add -json for full dependency details)
• tessl uninstall - remove tiles from your project

#What's Removed

The Framework functionality is no longer included. If you need the full framework features, v0.28.0 remains available (though development is paused).

Currently, automatic setup only supports Claude Code and Cursor. We'll add other agents based on requests—in the meantime, you can manually configure the MCP with any agent.

Note: This is a major rewrite (v0.30.0 → v0.50.2). Existing workflows will need adjustment.

(Updated Nov 27, 2025)

2025-10-17

#0.28.0

#New Features

• Invoking setup agent via npx will now configure mcp to use tessl via npx, ensuring smoother setup for users without a global install.
• Added integration with Cursor IDE for automatic generation and application of steering rules from tiles.
• Added tessl registry uninstall to easily uninstall tiles.

#Improvements

• Improved error messages for the add-member command to enhance clarity.
• Improved tessl status output to better reflect post-generation changes.
• Aligned arguments for tessl registry unpublish and tessl registry archive.

#Bug Fixes

• Fixed detection of direct dependencies in the pip detector.
• Resolved issue where unsupported package managers showed unnecessary user-facing messages.
• Extended display name support to include PyPI and Maven packages.
• Removed outdated compatibility code for specs.json.

2025-10-10

#0.27.0

#New Features

• New tessl setup project-info Command: CLI-only users can use this command to add details that Tessl tools require to AGENTS.md.
• Framework Initialization: you can choose “full-framework” (closed beta, login required) or “registry only” when running init.
• Steering Rules: Tiles can now define agent behavior rules via a new steering section in their manifests. These rules are automatically aggregated into a RULES.md file.
• Local Usage Spec Installation: Added support for installing tiles from local directories via a new source field in tessl.json.

#Improvements

• Simplified usage spec installation paths by removing version info from the directory structure.

#Bug Fixes

• Fixed config list to return empty results instead of throwing errors when no config file is present.

2025-10-07

#0.26.0

#New Features

• Selecting Cursor during tessl init or via tessl setup agent will now automatically install rule files into .cursor/rules/tessl/framework as part of project setup, improving Framework adherence.

2025-10-06

#0.25.0

#New Features

• Improved Tile Installation UX: Tiles suggested during install/update are now selected by default, making the default path simpler for users.
• Improved Help for tessl init: The -help output for tessl init now lists available agents, making it easier to discover options.
• Non-Interactive Shell Support: CLI installation no longer blocks in non-interactive shell environments.

#Improvements

• Better Logging Behavior: In non-interactive mode, JSON logs are now correctly sent to stderr, preserving standard piping behavior.
• Help Text Visibility: Tile help text remains visible even when tiles are deselected, improving clarity during installation.

#Bug Fixes

• Maven Dependency Detection: Fixed issues with the Maven dependency detector to ensure accurate detection.
• Logging Flushing Reliability: Addressed a flakiness issue by ensuring that stdio is flushed on close, not exit.
• Ink Render Handling: Corrected usage of Ink’s rendering logic to prevent hanging or exit-related issues.
• Security: Updated pino logger to remove a transitive vulnerability.

2025-10-02

#0.24.0

#New Features

• tessl registry sync to synchronize tiles from the registry.
• tessl setup framework to switch between the full Tessl framework and “registry-only” mode.
• tessl stamp to manually sync spec/code hashes after intentional edits so tessl status stops flagging them.
• tessl registry search now matches by full tile name (workspace/name), making it easier to find your own tiles.
• tessl registry list added to browse available tiles.
• tessl document adds tests to spec by default.

#Improvements

• tessl status surfaces an explicit error if bootstrap hasn’t been run.
• Tool calls will fail if agent bootstrap process has not been run, with guidance to fix.
• Added --print-project-cache flag for getting the path to the project cache.
• Removed interactive tool picker removed to simplify workflows.
• AGENTS.md now includes clearer “bootstrap” hints and wording tweaks to reduce confusion about when/why bootstrap should run.

#Bug Fixes

• MCP commands now authenticate correctly in all call sites.
• Reduced noisy output during tessl registry sync.

#Compatibility

• Minimum Node.js version is now >= 22.0.0.

2025-09-24

#0.23.0

#New Features

• Added support for GitHub Copilot.
• tessl edit now supports code files that have an associated @describe spec. Changes are made via the spec.

#Improvements

• tessl whoami command now includes your user ID.
• Reduced the chances of "token expired" errors, by proactively refreshing tokens when commands are run.
• Test commands will now time out if no output is received for a set period, preventing hangs.
• Registry commands now prompt for initialization if the workspace isn’t already set up.
• Unused .tessl/usage-specs directories are now removed during sync if the corresponding entries are removed from tessl.json.
• Test generation is now controlled by a unified generateTests param, which always skips locked tests.
• Improved list command usability, to make it easier for agents to select the right parameters.
• tessl registry publish skips uploading irrelevant .md and .gitignored files.

#Bug Fixes

• tessl status no longer shows duplicated test paths.
• Resolved issue where interactively selected tools weren’t able to run.
• Fixed Goose setup, so that correct .gooserules configuration is used.

#Deprecations

• Removed deprecated annotations like @describes, @generates, and @satisfied-by.

2025-09-19

#0.22.1

#Bug Fixes

• Fixed a project language detection issue that would occasionally cause errors across different commands.

2025-09-18

#0.22.0

#New Features

• init now offers the option to set up an agent and MCP integration when initializing a project for the first time.

#Improvements

• Support information is now listed in help for logged-out users.
• Tool parameters are now displayed before model parameters in tool-level help.
• Links are now supported in kitty terminal.

#Bug Fixes

• Fixed the rules file targeting in Gemini CLI agent setup.
• feedback is now gated to authenticated users, unauthenticated users can request support via email (support@tessl.io).

2025-09-16

#0.21.2

#Bug Fixes

• Fixed MCP server not allowing invocation of install or search tools when not logged in.

2025-09-16

#0.21.1

#Improvements

• Running tools with spec-registry-only context will now fail early and remind you to install the Tessl Framework before continuing.
• Added ability to exit registry search with the Escape key.
• Search results now link to the spec registry page instead of individual package registries.

#Bug Fixes

• Fixed Ctrl+C handling with the experimental observe command.
• Fixed a deprecation warning on registry install.

2025-09-15

#0.21.0

npm uninstall -g tessl-alpha
npm install -g @tessl/cli

#Breaking Changes

• Removed test generation from the build tool and introduced a new build-tests tool
• The spec format has been updated to allow capabilities to contain requirements that don’t have [@test] links:
  • create and edit specs will only add [@test] links if explicitly asked to do so in the --prompt

#New Features

• Our spec format now supports distinguishing between three types of tests:
  • locked tests: capture functional requirements provided
    • Denoted by { .locked } next to the [@test] link or in a header containing [@test] links.
    • You can manually mark tests as locked or use tessl test to do it by passing --lock-tests.
  • impl tests: cover implementation details that are tested to identify changes in behaviour disconnected from a spec’s functional requirements.
    • Denoted by { .impl }.
    • They are generated by tessl document --include-impl-details .
  • draft tests: are untagged and could have been added manually or using edit or create
    • You can use tessl test --lock-tests to automatically lock all draft the tests from a spec when all tests pass.
• Added flags to build-tests tool to select which kind of tests to generate --generate-draft-tests, --generate-locked-tests, and --generate-impl-tests.

#New Commands & Tools

• Added new registry install, registry search, and registry sync commands for working with Usage Specs.
• Added a new test tool, to run all of the tests for a given spec.
• Added generate-tests tool to add tests to a given spec.
• Added experimental observe tool to observe MCP tool invocations.

#Spec Registry

• Unauthenticated users can now access a subset of Tessl’s features to add Usage Specs to their projects.
  • Running init while logged out will set up AGENTS.md and framework files relating to the Tessl Spec Registry.

#Global Config

• config commands now support the --global flag, for updating the global Tessl config.
• The global config now supports a shareUsageData key to allow for analytics opt-out.

#Improvements

• Cleaner output for build and build-tests with a final summary listing successful and failed specs.
• Spec discovery for build is smarter: only builds specs with @generate links that actually need code updates.
• Diagnostics suggestions now correctly use positional parameters (e.g., tessl build <spec>), and status accepts <spec> as positional.
• Registry-only flows link to the Knowledge Index; usage spec manifest split into tessl.json (structured) and KNOWLEDGE.md (narrative).
• Tessl MCP tools can now log detailed output to a file to reduce their output and avoid saturating agent context when the agent is instructed to do so.
• Added --code and --tests filters to list command for finer control.
• Made test tools spec argument positional in CLI commands.
• status now splits diagnosed issues into errors and warnings and includes errors regarding spec parsing.
• config set now validates keys before setting.
• Updated spec syntax information in our system prompts:
  • Added context about our content tags: untagged (draft), locked and impl.
  • Removed un-used .dependencies and .dependency tags.
• Updated Tessl framework content with clearer guidance on tessl test, status, and fix behavior.
• version output is now cleaner and includes a hint to update if a new version is available.
• The --spec argument in build is now positional.
• Exposed modelName parameter in MCP tools

#Bug Fixes

• Spec selection no longer appears to “infinite scroll.”
• infer-spec no longer swallows errors.
• Fix loop now receives test results properly.
• Reporter failures print helpful context (e.g., missing file paths) to the terminal.
• Prevent generation of tests when using edit on code files.
• Test-runner name generation validates input and throws on invalid paths.
• Tessl framework files are now filtered from context provided to tools to prevent context poisoning.
• Fixed issue where login could not be cancelled while polling.
• Fixed truncated console logs when running in non-interactive/piped mode.
• Fixed edit tool so context files are properly passed.
• Fixed dependency verification to use correct @use path resolution.
• Prevented whitespace-only messages from being sent to models.
• Pass test results through fix loop iterations correctly.
• Fixed an incorrect path in .tessl/.gitignore for session data storage.
• Fixed init missing from global help.
• Reject invisible characters from LLM interactions.

2025-09-12

#0.19.0

#New Features

• Adopted AGENTS.md standard (replacing TESSL.md) for unified AI agent configuration across your development workflow.
• Improved Dependency Handling
  • verify-deps now runs automatically after code generation in the build process.
  • New specifyMissingDependencies flag can auto-update your spec with missing dependencies in tessl build and tessl edit.
  • New never option added to generation mode to skip file generation entirely.
• tessl status enhancements:
  • Highlights whether a spec includes @generate or @describe annotations.
  • Added dirtiness diagnostic to detect out-of-sync code/test files.
  • Added diagnostic for missing AGENTS.md entries in tessl status.
• Improved Spec & Test Management
  • Rewritten document tool with more intuitive behavior and support for complex specs.
  • Re-initializing a project now regenerates framework files.

#Improvements

• Removes setup language command; tools that require test commands now have a --test-command flag.
• Removed description upon tool invocation.
• Increased latest version check timeout in update command to reduce false negatives.

#Bug Fixes

• Logging in from an agent now automatically updates the MCP server so you can continue without restarting.
• Fixed tool summaries not showing in CLI output.
• CLI session data no longer prevents process exit after command completion.

2025-09-10

#0.18.1

#New Features

• build tool now has a post-generation fix loop to automatically resolve issues - we will be tweaking how we decide which test command to run in the coming week
• New status checks for orphaned tests, mixed implementation links & missing required files
• Added support for agent-native planning: agents now create and maintain .plan.md files automatically during planning.

#Improvements

• Removed --recursive option from status command for consistency with other commands
• Replaced --rebuild-tests with --generate-tests to build to generate only missing, outdated or always (default is now outdated rather than missing)
• Added --generate-code to build which behaves like --generate-tests but for code files (default is outdated, but is likely to change)
• Generated code now includes self-hashing for integrity checking and better status reporting on modified files
• Removed timestamps from generated code tags
• Improved logging and error messages across tools
• Improved exception messages to include more debugging context
• Better editor detection and logging in planning workflows
• Clarified descriptions for several commands & tools
• Tools invoked via MCP now return more information about their run, improving feedback visibility for agents
• The build tool now returns a failure exit code when the fix loop fails, ensuring accurate signaling of build status.

#Bug Fixes

• Fixed update command in Linux when using npm
• Fixed test generation messages saying “tests” instead of “test files”

#Deprecations

• Removed ability to edit plan files directly via the edit tool; users now receive a clear error if they attempt to do so.
• Made the plan tool private to reduce confusion and encourage agent-driven planning workflows

2025-09-09

#0.17.1

#Bug Fixes

• Fixes creation of the Claude subagent instructions directory as part of setup agent

2025-09-08

#0.17.0

#New Features

• Enabled tessl status to list all specs by default, providing broader project insights.
• Updated tessl status to include sync checks that detect whether a spec's code and tests are up to date.

#Improvements

• Updated build and document tool descriptions for clarity and accuracy.
• Made the -spec parameter more explicit and consistently required in tessl status help output.
• Renamed build tool's iterations parameter to maxIterations for clarity.
• Renamed @build links to @generate to reflect intended behavior.

#Bug Fixes

• Fixed bash execution in the update command to work reliably across Mac and Linux.
• Corrected error handling for expired keys.
• Made terms and conditions check respect a global directory set via environment variable.
• Fixed a crash in tessl status when a non-existent or empty spec file is provided.
• Resolved incorrect logging of issue counts in single-spec status output.