auto-complete-task
Skill3 starsChanged 39 days agoMITView on GitHub
Completes requested work autonomously through investigation, subtask planning, implementation, testing, and GitHub PR creation. Use when the user explicitly invokes auto-complete-task to complete a feature, bug fix, optimization, GitHub issue, or plain-text request.
In this repository12 other files it gives its agents.
- AGENTS.mdAGENTS.md
- audit-project-rulesSkill
- changelogSkill
- clear-writingSkill
- do-subtasksSkill
- investigate-issueSkill
- pest-arch-reviewSkill
- qaSkill
- retroSkill
- simplify-agent-instructionsSkill
- to-prSkill
- to-subtasksSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
25 rules. About 0 are common, 23 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈23
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Distinctive: line 22
Run this skill only when the user explicitly invokes it.
- Distinctive: line 23
Do not invoke this skill automatically from another skill or agent.
- Distinctive: line 24
Treat invocation as authorization to implement, test, commit, push, and create a PR for the request.
- Distinctive: line 25
Do not merge the PR.
- Distinctive: line 26
Follow repository instructions, tool permissions, and higher-priority restrictions.
- Distinctive: line 33
Treat features, bugs, and optimizations as valid requests.
- Distinctive: line 34
Ask what to work on only when no request was supplied.
- Distinctive: line 35
Read applicable repository instructions.
- Distinctive: line 36
Check the working tree with git status --short.
- Distinctive: line 38
Stop on dirty files unless the user explicitly authorized continuing despite them.
- Distinctive: line 39
Never modify, revert, or stage unrelated changes.
- Distinctive: line 41
Never create another worktree.
- Very specific: line 45
Default the target branch to main on origin.
- Distinctive: line 46
Use another target branch or remote when the user explicitly specifies one.
- Distinctive: line 51
Stop if the target is missing or fetching fails.
- Distinctive: line 52
Never substitute stale local branches or the current branch for the selected target.
- Very specific: line 56
Start a read-only subagent and instruct it to load investigate-issue.
- Distinctive: line 59
Use read-only Git commands to inspect target files when the current checkout differs.
- Distinctive: line 61
Do not authorize file edits or implementation in the investigation subagent.
- Distinctive: line 69
Ask only when missing access, conflicting requirements, or consequential ambiguity prevents safe progress.
- Distinctive: line 70
Do not invent requirements to avoid asking a necessary question.
- Distinctive: line 73
Use its autonomous handoff instead of requesting subtask approval.
- Distinctive: line 84
Follow its implementation, testing, separate-commit, push, and PR verification rules.
- Distinctive: line 86
Do not push partial work or create a PR before all subtasks and required checks pass.
- Distinctive: line 88
If blocked, report completed work and the exact action needed to continue.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Auto Complete Task
Quick start
Example: Auto-complete-task: fix the search timeout described in #123.
- Accept the work request.
- Delegate investigation to a subagent using
investigate-issue. - Load
to-subtasksand generate implementation subtasks. - Load
to-prand execute the subtasks without separate approval. - Return the PR URL and validation results.
Authorization
- Run this skill only when the user explicitly invokes it.
- Do not invoke this skill automatically from another skill or agent.
- Treat invocation as authorization to implement, test, commit, push, and create a PR for the request.
- Do not merge the PR.
- Follow repository instructions, tool permissions, and higher-priority restrictions.
Workflow
1) Intake and preflight
- Accept a GitHub issue URL, issue number, or plain-text work request.
- Treat features, bugs, and optimizations as valid requests.
- Ask what to work on only when no request was supplied.
- Read applicable repository instructions.
- Check the working tree with
git status --short. - Preserve the dirty-file safeguards from
to-pr. - Stop on dirty files unless the user explicitly authorized continuing despite them.
- Never modify, revert, or stage unrelated changes.
- Stay in the current directory, including when it is already a worktree.
- Never create another worktree.
2) Resolve the target
- Default the target branch to
mainonorigin. - Use another target branch or remote when the user explicitly specifies one.
- Fetch the selected target before investigation.
- For the default, run
git fetch origin main. - Resolve the fetched target commit with
git rev-parse FETCH_HEAD. - Record the target branch name, remote, and fetched commit separately.
- Stop if the target is missing or fetching fails.
- Never substitute stale local branches or the current branch for the selected target.
3) Investigate in a subagent
- Start a read-only subagent and instruct it to load
investigate-issue. - Supply the request, repository path, target commit, constraints, and relevant conversation context.
- Require investigation against the fetched target commit.
- Use read-only Git commands to inspect target files when the current checkout differs.
- Require evidence, acceptance criteria, proposed changes, validation options, and blockers.
- Do not authorize file edits or implementation in the investigation subagent.
- Continue factual investigation when findings are insufficient.
4) Plan autonomously
- Resolve routine uncertainty from repository evidence and established project conventions.
- Choose the smallest supported change that satisfies the request.
- Record reasonable, non-blocking assumptions.
- Ask only when missing access, conflicting requirements, or consequential ambiguity prevents safe progress.
- Do not invent requirements to avoid asking a necessary question.
- Load
to-subtaskswith the request, investigation findings, acceptance criteria, and constraints. - Identify this as an explicitly user-invoked
auto-complete-taskworkflow. - Use its autonomous handoff instead of requesting subtask approval.
- Include relevant regression coverage and validation checks in the implementation subtasks.
5) Implement and open the PR
- Load
to-prwith the authorized subtasks. - Explicitly supply the fetched commit as the work branch starting point.
- Explicitly supply the selected branch name as the PR base, such as
main, notorigin/main. - Supply the selected remote and original GitHub issue reference when applicable.
- Apply these branch instructions only to this run.
- Preserve the standalone current-branch default in
to-pr. - Follow its implementation, testing, separate-commit, push, and PR verification rules.
- Preserve its safeguards, including dirty-file, label-selection, and validation blockers.
- Do not push partial work or create a PR before all subtasks and required checks pass.
- Return the PR URL, completed work, validation results, assumptions, and blast-radius rationale.
- If blocked, report completed work and the exact action needed to continue.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

