agentleFS
seancabalseseancabalse/agentic-playbook/skills/deliver-feature/SKILL.md

deliver-feature

Skill0 starsChanged 15 days agoMITView on GitHub

Deliver a ticket from start to end with the agents: brief, plan, build, verification, reviews, fixes and docs. It stops before shipping. Run only when the user asks.

3 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read the whole diff yourself, in its worktree (git -C <path> diff <feature-branch>...HEAD).

    line 37
  2. Distinctive:

    Run the full check suite yourself, in the worktree.

    line 38
  3. Distinctive:

    Check the diff against the plan and the scope.

    line 39
  4. Distinctive:

    Read its handoff, and re-check one key claim yourself (for example, open one screenshot).

    line 52
  5. Very specific:

    Start these together, each with the diff (git diff <base>...HEAD), the ticket and the qa handoff:

    line 56
  6. Distinctive:

    security-auditor: when auth, data access, files, payments or personal data changed.

    line 58
  7. Distinctive:

    design-reviewer: when UI changed; pass it the screenshot paths.

    line 59
  8. Very specific:

    Make sure nothing is left running.

    line 78

≈ “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

Deliver a feature with handoffs

Playbook files named here: when the project has its own copy at .playbook/local/<same path>, use that copy instead.

Ticket: $ARGUMENTS

You are the orchestrator, the tech lead. You plan, delegate, review and verify. The product-engineer agent writes the code, one writer at a time; you only write small changes yourself. Every agent returns a handoff (${CLAUDE_PLUGIN_ROOT}/templates/handoff.md), and you check its key claims before acting on it. Rules: ${CLAUDE_PLUGIN_ROOT}/principles/15-agent-collaboration.md.

1. Plan (gate: owner)

If the request isn't clear enough to plan (goal, users, scope or acceptance criteria missing), delegate to the product-owner agent first. Ask the owner its blocking questions, run it again with the answers, and repeat until it returns a brief ready to agree; the owner agrees it, and you post it on the ticket. Then follow start-task. Post the plan and assumptions on the ticket. If there are product questions, ask and stop until the owner answers. Move the ticket to the tracker's in-progress status.

2. Design (only for structural work: new tables, flows, vendors, sync)

Delegate to the architect agent with the ticket, the plan and the relevant code paths. If its handoff says an ADR is needed, write it (write-adr), then ask the owner to accept it and stop until they do.

3. Build

Delegate to the product-engineer agent. Give it:

  • the ticket and acceptance criteria;
  • the owner's decisions;
  • the plan with its checks;
  • the architect's design, if there was one;
  • the files and area it may change.

It follows implement (preloaded) and works in its own worktree (isolation: worktree), on a branch from your feature branch. Its handoff names the worktree path and branch.

  • If it returns blocked with a product question, ask the owner, then send the answer back (continue the same agent if your tool allows, or start a new run with its handoff).
  • When it returns done:
    • Read the whole diff yourself, in its worktree (git -C <path> diff <feature-branch>...HEAD).
    • Run the full check suite yourself, in the worktree. Its "passed" is a claim, not proof.
    • Check the diff against the plan and the scope.
    • Then bring it into the feature branch (merge or cherry-pick its commits) and remove the worktree. Don't push until the owner ships.
  • For a change of a few lines, build it yourself following implement, rather than delegating.
  • Independent slices (no shared files) may go to separate product-engineer runs, each in its own worktree. Otherwise run them one after another.

4. Verify

Delegate to the qa-verifier agent with:

  • the acceptance criteria;
  • which states and settings matter;
  • the screenshot matrix (capture-screenshots);
  • the runbook's launch steps.

Read its handoff, and re-check one key claim yourself (for example, open one screenshot).

5. Review, in parallel, read-only

Start these together, each with the diff (git diff <base>...HEAD), the ticket and the qa handoff:

  • code-reviewer: always.
  • security-auditor: when auth, data access, files, payments or personal data changed.
  • design-reviewer: when UI changed; pass it the screenshot paths.

6. Fix (at most 2 loops)

Decide each finding first (principle 13, "Fix now, or ask the owner"): ask the owner about the ones that would widen the scope. Send the other confirmed findings (and the plausible ones you've checked) to the product-engineer, with the reviewers' scenarios and any failing tests they wrote in their worktrees. It fixes them in a fresh worktree from the feature branch. When it returns, re-read the diff, re-run the full checks, then have qa-verifier re-check only what changed. After two loops, stop and report what's left instead of looping.

7. Docs

Delegate to the docs-keeper agent with the diff, the decisions made and the handoffs. Check its list of updated docs.

8. Stop for the owner

Post a summary handoff on the ticket. It covers:

  • what was built;
  • the evidence: checks, screenshots, the review results;
  • what wasn't verified;
  • the decisions made;
  • what's next.

Make sure nothing is left running. Then tell the owner it's ready, and that /ship-pr opens the PR. Don't commit, push or open the PR yourself unless the owner says so in this conversation.

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.