agentleFS
AIPrisAIPris/ticket-to-merge/skills/issue-loop/SKILL.md

issue-loop

Skill0 starsChanged 25 days agoMITView on GitHub

Autonomously work through all open GitHub issues of the current repo - a fresh Claude session per issue, an independent code review, then commit, PR, CI wait/repair, squash merge. Only run when the user explicitly invokes /issue-loop.

4 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions4 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 4

    1. deedeeharris/claude-skills★ 7same name
    2. ksprashu/agent-skill-forge★ 3same name
    3. timmo001/skills★ 2same name
    4. musoukun/skills★ 1same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 4 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 4

    Distinctive

    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
  1. Distinctive:

    Usage limit: if a Claude session hits the usage limit (5-hour or weekly), the supervisor does not stop.

    line 24
  2. Distinctive:

    Run the read-only preflight and show the user the output (repo, base branch, verify commands, next issue):

    line 31
  3. Distinctive:

    If it fails (dirty tree, unpushed commits on the base branch, wrong branch, missing tool, no auth), explain the cause and stop.

    line 37
  4. Distinctive:

    After confirmation start it in the background so the session stays usable:

    line 45
  5. Very specific:

    Use run_in_background. To report progress, run the quick status check (read-only, safe while the loop runs):

    line 51
  6. Very specific:

    When it finishes, summarize: issues merged, issues parked (with their PR links), or the exact stop reason and where unfinished work was …

    line 63

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Issue loop (global, Windows)

Runs ${CLAUDE_SKILL_DIR}/run_issues.py against the git repository in the current working directory. The script is a Python supervisor: for each open issue (oldest first) it starts a fresh claude -p worker, verifies locally, commits, has the commits reviewed, pushes agent/issue-N, opens a PR (Closes #N), waits for CI (fresh repair sessions on red CI), squash-merges, deletes the branch, then continues. It stops on any other failure and never bypasses branch protection.

Code review gate (before every push, including pushes of CI repairs): a fresh read-only reviewer session checks the commits. It blocks only on bugs, unsolved issue requirements, missing tests and broken CLAUDE.md/AGENTS.md rules. On findings, a fresh fixer session fixes or rebuts each one, then a narrow re-review follows (default: up to 2 fix rounds). Each approval is posted on the PR as a comment. If the last re-review still requests changes, the issue is parked, not merged: branch and PR stay open with the findings as a PR comment, the issue gets the label agent-needs-human, and the loop moves on to the next issue. Parked issues are skipped until someone removes the label.

Usage limit: if a Claude session hits the usage limit (5-hour or weekly), the supervisor does not stop. It sleeps until the reset time Claude reports (plus 2 minutes), then re-runs that session fresh; a writing session is told that the tree may hold the interrupted attempt's partial work. --status shows the resume time. A weekly limit can mean a wait of days. Other API errors (overloaded, 5xx) still stop the loop.

Procedure

  1. Run the read-only preflight and show the user the output (repo, base branch, verify commands, next issue):

    python "${CLAUDE_SKILL_DIR}/run_issues.py" --repo . --check

    If it fails (dirty tree, unpushed commits on the base branch, wrong branch, missing tool, no auth), explain the cause and stop. Do not "fix" the user's working tree yourself.

  2. State clearly: the workers run with --dangerously-skip-permissions, and merges into the default branch happen automatically. Ask the user to confirm before starting (use the scope in $ARGUMENTS, e.g. --max-issues 1; if none was given, run without a limit, i.e. until all open issues are solved. Say so in the confirmation; do not suggest or add --max-issues on your own).
  3. After confirmation start it in the background so the session stays usable:

    python "${CLAUDE_SKILL_DIR}/run_issues.py" --repo . $ARGUMENTS

    Use run_in_background. To report progress, run the quick status check (read-only, safe while the loop runs):

    python "${CLAUDE_SKILL_DIR}/run_issues.py" --repo . --status

    It prints RUNNING / not running, the current issue, the stage (Claude working, verifying, waiting for CI, merging, ...), how many issues were merged this run, and seconds since the last activity (flags >10 min as possibly stuck). Prefer this over reading the raw output file. State lives in %LOCALAPPDATA%\issue-loop\<repo>.status.json. While it runs, do not touch the repository (no edits, checkouts or commits) - the supervisor needs a clean tree.

  4. When it finishes, summarize: issues merged, issues parked (with their PR links), or the exact stop reason and where unfinished work was kept (agent/failed-issue-N-* branch, or an open PR).

Per-repo settings

Optional <repo>\.issue-loop.json (commit it - the tree must stay clean):

{ "verify": ["npm run build", "npm test"], "protected_paths": ["vendor/"], "model": "opus",
  "worker_model": "sonnet", "review_model": "opus",
  "worker_max_turns": 200, "ci_repair_max_turns": 100, "ci_repair_attempts": 3, "local_repair_attempts": 2,
  "review": true, "review_rounds": 2, "review_max_turns": 80, "allow_no_ci": false }

model applies to every session; worker_model (sessions that change code) and review_model (review sessions) override it. "review": false turns the review gate off. review_rounds is the number of fix + re-review rounds before an issue is parked. Counts must be whole numbers (0 or more; *_max_turns 1 or more), otherwise the preflight fails. If a workflow runs on pull requests but GitHub reports no checks for the PR, the loop stops and leaves the PR open; "allow_no_ci": true merges in that case instead (e.g. path-filtered workflows).

Without verify, commands are auto-detected (package.json build/test scripts, or pytest). Logs and lock files live in %LOCALAPPDATA%\issue-loop\.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.