agentleFS
Koushik890Koushik890/ship-task/SKILL.md

ship-task

Skill0 starsChanged 36 days agoMITView on GitHub

Run the full michaelshimeles workflow on one task, start to PR - isolate in a worktree, build, prove with evidence, open a PR with before/after proof, then loop Greptile to 5/5. Invoke with /ship-task <task description>.

  • Commits and pushes

1 other project uses a version of this.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 1

    1. aiatelie/ai-atelie★ 9same 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

8 rules. About 0 are common, 3 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If the task is empty or too vague to test, ask one short question before starting.

    line 12
  2. Very specific:

    Must be inside a git repo with a GitHub origin and an authenticated gh (gh auth status).

    line 18
  3. Very specific:

    Start the task in its own worktree branched from latest origin/main (or the repo's default branch), including the scope check against open …

    line 27
  4. Distinctive:

    Run the repo's checks from preflight.

    line 36
  5. Very specific:

    Commit with a clear message, rebase onto latest origin/<default branch>, rerun checks, push (git push -u origin <branch>.

    line 41
  6. Very specific:

    Write the PR body: what changed, how it was tested (every claim backed by evidence), risks and follow-ups.

    line 42
  7. Very specific:

    Run /greploop until Greptile gives 5/5 with zero unresolved comments (use /greploop-apps if Greptile says too many files changed).

    line 44
  8. Distinctive:

    Keep the worktree until the PR is merged or closed.

    line 49

≈ “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.

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

Ship Task

Task: $ARGUMENTS

If the task is empty or too vague to test, ask one short question before starting.

Run the four beats below in order, invoking each named skill with the Skill tool. Do not skip a beat silently: if a beat cannot run (missing tool, no GUI, no Greptile), say so in one line, use the fallback, and keep going. Keep a short checklist of beats and tick each one off as you go.

0. Preflight

  • Must be inside a git repo with a GitHub origin and an authenticated gh (gh auth status). If not, stop and tell the user what is missing.
  • The skills this workflow calls must be installed: new-feature, code-structure, evidence-driven-testing, unslop, greploop, greploop-apps, before-and-after. If any is missing from your skill list, stop and tell the user to run the installer from https://github.com/Koushik890/ship-task. Do not improvise a replacement for a missing skill. The before-and-after skill must be the vercel-labs version: its directory contains scripts/format.mjs. If it has scripts/capture.sh instead, it is the older copy vendored in michaelshimeles/skills, which this workflow does not support; tell the user to rerun the installer.
  • Find the default branch with gh repo view --json defaultBranchRef -q .defaultBranchRef.name (fall back to main). Wherever this skill says origin/main, use origin/<default branch>.
  • Check optional tools once and remember which are missing: jq, node, ffmpeg, agent-browser, and Python. On Windows python3 is often absent or a Microsoft Store stub; try python3, then python, then py -3, and use whichever prints a 3.x version for every Python command the other skills give you.
  • Shell: write commands for the shell tool you are actually running (bash/Git Bash vs PowerShell). Use forward slashes and quote paths that can contain spaces.
  • Find the repo's checks: CLAUDE.md / AGENTS.md first, then package.json scripts (test, lint, typecheck, build), Makefile, pyproject.toml, etc. Note the exact commands; you will run them in beat 3 and before pushing.

1. Isolate - /new-feature

Start the task in its own worktree branched from latest origin/main (or the repo's default branch), including the scope check against open PRs. On overlap with another open PR, stop and ask.

2. Build - /code-structure

Implement the smallest change that does the task. Follow code-structure: actions own the why/when, a service layer owns reusable how. Do not refactor beyond what the task needs.

3. Prove - /evidence-driven-testing

  • Capture the before state first, while reproducing the issue or showing the missing feature, before the fix lands where possible.
  • Run the repo's checks from preflight; all must pass.
  • Gather runtime evidence of the after state. UI change: screen recording if a GUI is drivable, else the skill's headless path (Playwright screenshots). Non-UI change: measured numbers or before/after output pairs.

4. Ship - /before-and-after, /unslop, /greploop

  1. Commit with a clear message, rebase onto latest origin/<default branch>, rerun checks, push (git push -u origin <branch>; --force-with-lease only on your own branch, never plain --force).
  2. Write the PR body: what changed, how it was tested (every claim backed by evidence), risks and follow-ups. Run /unslop over the PR title, body and commit messages, then open the PR with gh pr create.
  3. Visible change: capture before/after screenshots or recordings with agent-browser (run agent-browser skills get core --full for usage) into a captures/ folder, then run /before-and-after to insert the comparison block into the PR. Its formatter is scripts/format.mjs inside the before-and-after skill's own directory (the skill's base directory is shown when it loads; do not assume ~/.claude/skills), run with node; it uploads through gh --attach (GitHub-hosted, needs gh 2.100+). Never commit captures/: add captures/ to .git/info/exclude before capturing.
  4. Run /greploop until Greptile gives 5/5 with zero unresolved comments (use /greploop-apps if Greptile says too many files changed). If Greptile is not installed on the repo or never responds, say so and skip this step.

Finish

  • Do not merge the PR unless the user explicitly says to.
  • Keep the worktree until the PR is merged or closed.
  • End with: PR URL, one line per beat (done / skipped and why), and the Greptile score.

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.