agentleFS
ryanmoelterryanmoelter/plume/.claude/skills/implement/SKILL.md

implement

Skill0 starsChanged 10 days agoApache-2.0View on GitHub

Implement one Plume roadmap item end to end — a Linear issue ID, a Linear URL, or a described task — in its own worktree, with fresh-agent code reviews, a debug build for the user's manual review, and then a hand-off to the release agent, a merge into main, or a PR. Use on "/implement PLUME-123", "work on PLUME-123", "implement this issue", or a task description meant to land as one change.

191 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use setup-ts-deep-modules, ask-matt and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • liufeng1976Skill

    setup-ts-deep-modules

    Wire dependency-cruiser into a TypeScript repo so each package is a deep module — implementation hidden in subfolders, reachable only through its entry-point files. User-invoked.

    Also in 43 projects that use this

  • GoogleSkill

    ask-matt

    Ask which skill or flow fits your situation. A router over the skills in this repo.

    Also in 91 projects that use this

  • RoentekSkill

    loop-me

    Grill me about specs for the workflows I want to build, within this workspace.

    Also in 50 projects that use this

  • AWSSkill

    to-questionnaire

    Turn a decision you can't fully answer into a questionnaire for someone else to fill in.

    Also in 88 projects that use this

  • ryotaroh180105Skill

    claude-handoff

    Hand the current conversation off to a fresh background agent that picks up the work immediately.

    Also in 42 projects that use this

  • GoogleSkill

    to-spec

    Turn the current conversation into a spec and publish it to the project issue tracker — no interview, just synthesis of what you've already discussed.

    Also in 168 projects that use this

  • +3 more from google

The file

Show its rules and text

We split this skill into the rules it gives an agent

19 rules. 0 are common, 7 are distinctive, and 12 are very specific to this project.

  • 0

    Common

    Most of the 171 edited versions we read keep these.

  • 7

    Distinctive

    This author's own choices. Worth a look.

  • 12

    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
  1. Very specific:

    If the argument is a PLUME-<n> identifier or a linear.app/…/issue/PLUME-<n>… URL, fetch the issue with get_issue, plus its comments and …

    line 13
  2. Very specific:

    Create the worktree through the worktrees skill: wt cob <slug>-plume-<n> --no-tab, then EnterWorktree with the path from wt list.

    line 20
  3. Very specific:

    If the user names another base, such as the release branch, create the branch from it with git branch first, then wt co <branch> --no-tab.

    line 21
  4. Distinctive:

    Every read, edit, build and command stays inside the worktree.

    line 22
  5. Distinctive:

    Every sub-agent prompt carries:

    line 28
  6. Distinctive:

    Never run git stash; the stash stack is shared by every worktree.

    line 31
  7. Distinctive:

    Run builds and tests in the foreground, never through a background watcher.

    line 32
  8. Distinctive:

    No screenshots unless the prompt asks for one.

    line 35
  9. Very specific:

    Commit as you go, following the commit-message skill and the Commits section of CLAUDE.md, trailer included.

    line 37
  10. Very specific:

    Verify through a Sonnet agent: xcodebuild … build, then test -only-testing:PlumeTests, confirming test names scroll past.

    line 39
  11. Distinctive:

    Before the first debug build, review until clean:

    line 43
  12. Very specific:

    Only after the user approves.

    line 65
  13. Very specific:

    PR, only when the user asks for one: Use the open-pr skill. The issue stays In Progress.

    line 67
  14. Very specific:

    Release agent, if one is live: Run ListAgents and look for a session named like Release <version>.

    line 68
  15. Distinctive:

    Wait for its reply that the branch landed before cleaning up.

    line 76
  16. Very specific:

    Delete the branch once its content has landed: git diff <branch>..<target> --stat must be empty.

    line 82
  17. Very specific:

    Remove the worktree's debug data, ~/Library/Application Support/Plume.debug-<worktreeName>/.

    line 83
  18. Very specific:

    Remove the worktree's DerivedData.

    line 84
  19. Very specific:

    Never pkill Plume, pgrep -x Plume, or global-grep for claude.

    line 89

“Common” is measured: kept by at least 6 in 10 of the 171 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

Implement

Take one item from intake to landed. You are the coordinator: delegate exploring, implementing, building and reviewing to sub-agents, and keep your own context for decisions, commits, Linear and the user. The user's manual review is the one hard gate.

0. Intake

  • If the argument is a PLUME-<n> identifier or a linear.app/…/issue/PLUME-<n>… URL, fetch the issue with get_issue, plus its comments and attachments. Read the code for what already exists; the issue states only the item.
  • Otherwise the argument is the spec. Ask whether to create an issue for it; the answer is usually yes. On yes, create it through the roadmap skill in the project for its outcome, ask the user for a size, and carry on with the new identifier.
  • State your understanding in one paragraph. Ask now only if the issue leaves a load-bearing choice open; otherwise start without a plan gate.
  • Move the issue to In Progress through the roadmap skill. Record important decisions on the issue as comments when they are made.

1. Worktree

  • Create the worktree through the worktrees skill: wt cob <slug>-plume-<n> --no-tab, then EnterWorktree with the path from wt list. wt branches off main and adds the ryanm/ prefix.
  • If the user names another base, such as the release branch, create the branch from it with git branch first, then wt co <branch> --no-tab.
  • Every read, edit, build and command stays inside the worktree. The merge into main in step 6 is the only exception.

2. Implement

Delegate each step to a sub-agent. Set model explicitly: Sonnet for S-sized or mechanical work and for every build, test and launch; Opus for M or larger, unsized work, or anything touching the headless wire protocol.

Every sub-agent prompt carries:

  • The worktree's absolute path, and the rule to confine all reads, edits and commands to it.
  • Never run git stash; the stash stack is shared by every worktree.
  • Run builds and tests in the foreground, never through a background watcher.
  • Report any commit made with --no-gpg-sign.
  • distress-call exists; use it rather than guess on a decision only the user can make.
  • No screenshots unless the prompt asks for one.

Commit as you go, following the commit-message skill and the Commits section of CLAUDE.md, trailer included.

Verify through a Sonnet agent: xcodebuild … build, then test -only-testing:PlumeTests, confirming test names scroll past. CLAUDE.md lists the suites that fail for environmental reasons; anything else is yours to fix. Add tests for pure logic.

3. Review loop

Before the first debug build, review until clean:

  1. Spawn a fresh Opus reviewer. Give it only the worktree path, the diff range (main...HEAD, or the base branch), and the issue's requirements — not your reasoning. It runs /code-review and reports findings. Never review inline.
  2. Triage the findings yourself. Have a sub-agent fix the real ones, then re-test.
  3. Spawn a new reviewer and repeat. Stop when a round finds nothing real. After three rounds, surface what is left to the user instead of looping.

4. Debug build and manual review

A Sonnet verifier builds and launches the worktree's own bundle:

  • Find it with -showBuildSettings | awk '$1 == "BUILT_PRODUCTS_DIR"', and launch with env -u CLAUDE_CODE_CHILD_SESSION -u CLAUDECODE open <app>. Without the env -u, every agent tab's claude reads itself as nested and the chat shows nothing.
  • Quit or relaunch only PIDs matching this bundle path (pgrep -f "^<app>/Contents/MacOS/Plume"). Other agents run their own builds.
  • Confirm the launch with the drive-plume skill, then send command -v plume >/dev/null && plume notify "Ready to test" "PLUME-<n>".

Tell the user it is running and give a short checklist of what to try. Iterate on their feedback: a sub-agent edits, the verifier rebuilds and relaunches. Skip the review loop between iterations. Rebase onto main or the release branch when asked. If the user waives manual testing, note it and move on.

5. Final review

Once the user is happy, run one more fresh reviewer over the whole diff. Fix and re-test anything real; relaunch for a quick look if the fixes changed behavior.

6. Land

Only after the user approves. Check signatures first — git log --format='%G? %h %s' <base>..HEAD | grep -v '^G' must print nothing — and re-sign any offender while it is still one branch deep. Then take the first path that applies:

  1. PR, only when the user asks for one. Use the open-pr skill. The issue stays In Progress.
  2. Release agent, if one is live. Run ListAgents and look for a session named like Release <version>; ask if more than one could be it. Rebase onto its release branch if needed and re-test. Then SendMessage it a hand-off:

    • The issue identifier and title, and that the user approved it.
    • Branch, worktree path, commit SHAs, and the base it sits on.
    • Signing state, and verification results: build, tests, review rounds.
    • Known unrelated failures, verbatim.
    • Changelog bullets in the CHANGELOG.md voice.
    • "Leave PLUME-<n> open until it reaches main."

    Wait for its reply that the branch landed before cleaning up. The issue stays In Progress; the release agent moves it to Done.

  3. Otherwise, merge into main. Rebase onto main and re-run the tests. Then, in the primary checkout, confirm it is on main and clean — if it is not, stop and ask, because another agent may own it. Run git merge --no-ff -S <branch>, push, and confirm git rev-list --count main...origin/main is 0. Mark the issue Done through roadmap; the user's approval of the merge is the confirmation.

7. Clean up

  • ExitWorktree with action: keep, then wt delete <name>.
  • Delete the branch once its content has landed: git diff <branch>..<target> --stat must be empty. A re-signed cherry-pick reads as unmerged to git branch --merged.
  • Remove the worktree's debug data, ~/Library/Application Support/Plume.debug-<worktreeName>/.
  • Remove the worktree's DerivedData. Quit its debug app first (only the PID from pgrep -f "^<app>/Contents/MacOS/Plume"). The folder is the ~/Library/Developer/Xcode/DerivedData/Plume-* whose info.plist has WorkspacePath equal to <worktree>/Plume.xcodeproj; read it with /usr/libexec/PlistBuddy -c 'Print :WorkspacePath' <dir>/info.plist. Delete only an exact match, never the primary checkout's.
  • Report what landed and where. For anything deferred, ask before creating an issue, and take its fields from the user.

Gotchas

  • Never pkill Plume, pgrep -x Plume, or global-grep for claude. Walk down from your own build's PID.
  • A hidden chat tab reads as disabled and "not found" in drive-plume. Check which tab chip is selected before believing a reading.
  • An isolation: "worktree" agent forks from origin/main, not your branch. Have it run git checkout -B <branch-for-it> <your branch> first, or have it work in your worktree instead.

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.