linear-task-worker/SKILL.md
Autonomous Linear task worker that selects Linear issues, implements them with TDD, self-reviews, commits, pushes, and moves finished work to In Review.
npx skillsauth add sssemil/skills linear-task-workerInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Linear-backed autonomous task worker. Use this when the user asks for
$linear-task-worker, asks to work Linear tickets, or asks for task-worker
behavior without local workspace task files.
This skill preserves the implementation discipline of task-worker, but Linear
is the only persistent task state.
AGENTS.md.AGENTS.md must name the Linear project to use. Accept clear wording such as
Linear project: <name> or "Use the Linear project <name>".workspace/plans, workspace/tasks, or
workspace/review-state files.type:plan, type:task, type:review-finding.In Review, not directly to Done.AGENTS.md.TARGET.md if present.If uncommitted changes exist:
Prefer issues in this order:
In Progress.In Progress issue in the resolved project if the user explicitly asked
to continue active work.Todo issues in deterministic order, preferring lower Linear
identifier or oldest creation date.Only select issues labeled type:task or type:review-finding. Do not select
type:plan parent issues for implementation.
Skip issues that are blocked by another unresolved issue. If nothing is available, report that there is no unblocked Linear work.
After selection:
In Progress.Started work via linear-task-worker.Before editing:
AGENTS.md, issue body, or repo scriptsIf the issue is not decision-complete, comment with BLOCKED: explaining the
missing decision, move it to Backlog, and stop or pick the next unblocked
issue if the user requested continuous processing.
For each requirement:
Use repo commands from AGENTS.md first. If the repo has ./run, prefer it over
raw package-manager commands.
Before committing:
Run a self-review. Use subagents when useful and available. Review from these perspectives:
Categorize findings:
CRITICAL: must fix before completionMAJOR: must fix before completionMINOR: comment or fix if cheapNIT: optionalFix all CRITICAL and MAJOR findings, then re-run relevant verification.
When verification passes:
In Review.Do not move to Done unless the user explicitly says review/acceptance is
complete.
If the user asked to process multiple tasks:
In ReviewTodo or resumable In Progress issues remainIf repeated blockers prevent progress on the same issue, leave a precise
BLOCKED: comment and move on only when another unblocked issue exists.
Summarize:
Keep the response concise and factual.
tools
Systematically reviews a project subsystem-by-subsystem with resumable .brutal-workspace state and creates Linear review finding issues for CRITICAL and MAJOR problems.
development
Collaborative, multi-perspective feature planning with rigorous requirements interrogation. Creates Linear project documents and Linear issues instead of local workspace plan/task files.
documentation
Compact the current conversation into a handoff document for another agent to pick up.
development
Delegate coding work to OpenCode through safe-opencode. Use when Codex should plan, route the job to a suitable DeepSeek model, review git diff, and validate instead of writing the main code itself.