agentleFS
maxedappsmaxedapps/agent-skills/skills/implement-plan/SKILL.md

implement-plan

Skill59 starsChanged 52 days agoMITView on GitHub

Implements a saved plan through to a tested, independently reviewed pull request, handing it to an agent in its own worktree when possible. Use this skill when the user asks to implement, execute, or continue a plan. Do not use for creating a plan (use create-plan) or for changes without a plan.

28 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use iterate-plan, validate-plan and 8 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.
  • corticalstackSkill

    iterate-plan

    Revise an existing implementation plan in flow/plans/ based on feedback, researching when needed. Explicit-only; run only when invoked via /iterate-plan or explicitly asked.

    Also in 5 projects that use this

  • shawn-dumasSkill

    validate-plan

    Validate an orchestration plan after pre-flight certification. Runs adversarial review, deep review (verify prompt data against codebase), and PoC gate for risky approaches. Auto-invoked by orchestrate-* skills at Step 8.

    Also in 4 projects that use this

  • parcadeiSkill

    describe-pr

    Generate comprehensive PR descriptions following repository templates

    Also in 4 projects that use this

  • joshuaalpuertoSkill

    research-codebase

    Conducts structured research across a codebase to answer questions about architecture, components, and data flow. Use when investigating how code works, where functionality lives, or how systems interact. Decomposes questions, runs parallel investigations, synthesizes findings with file:line references, and produces persistent research documents.

    Also in 4 projects that use this

  • JetBrainsSkill

    review-plan

    Manually re-run the autonomous plan review (Phase 2 — consistency + structural). The same review runs automatically as the first phase of /execute-tracks; this command is for re-runs after inline replanning or whenever the plan needs explicit re-validation. TRIGGER when: user asks to re-validate the plan after inline replanning; user requests explicit plan review outside /execute-tracks. SKIP: /execute-tracks already running State 0 in this session (autonomous Phase 2 path will fire) — do not double-run.

    Also in 5 projects that use this

  • JetBrainsSkill

    fix-bug

    End-to-end bug fix workflow for the Exposed project. Accepts a GitHub issue (#NUMBER), YouTrack issue (EXPOSED-NUMBER), or YouTrack URL (https://youtrack.jetbrains.com/issue/EXPOSED-NUMBER). Fetches the issue, creates a failing reproducer test, commits on a new branch, implements the fix, validates, and opens a PR. Use this skill whenever the user wants to fix a bug from an issue tracker, mentions a EXPOSED issue number, or references a GitHub issue to fix.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

11 rules. 0 are common, 11 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 28 edited versions we read keep these.

  • 11

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Create a worktree (see use-worktrees).

    line 17
  2. Distinctive:

    When the review is settled, quit the reviewer and close its pane.

    line 21
  3. Distinctive:

    When the implementer reports its PR, check the branch is fully pushed, quit the agent, close its pane, remove the worktree (see …

    line 22
  4. Distinctive:

    Read the plan and its ADR.

    line 28
  5. Distinctive:

    Add or update tests only where they protect behavior that matters (see awesome-tests).

    line 31
  6. Distinctive:

    Verify every task yourself: run the checks the plan names, and check anything with a UI in a real browser with agent-browser.

    line 32
  7. Distinctive:

    Keep the plan file current.

    line 33
  8. Distinctive:

    When all tasks are done, get the review round.

    line 34
  9. Distinctive:

    Commit, push, and open a pull request that says what changed, how you verified it, and what the review changed.

    line 35
  10. Distinctive:

    Stop dev servers and close browsers.

    line 36
  11. Distinctive:

    Every implementation gets one review by an agent that didn't write it, to catch what the author can't see, not to grow the code.

    line 42

“Common” is measured: kept by at least 6 in 10 of the 28 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 Plan

Hand off

Inside herdr (HERDR_ENV=1), hand the plan to a new agent unless you are the implementer or were told to implement it here (see orchestrate-agents):

  1. Create a worktree (see use-worktrees). Copy the plan and ADR into it if they aren't committed.
  2. Open the worktree in a new pane of your current herdr workspace, tagged as your worker (see use-worktrees), and start an agent of your kind there. Never create a new workspace or tab for it. Give it the plan path, any context the plan lacks, and: "You are the implementer. Use implement-plan in this checkout; don't hand off."
  3. Step in only when it is blocked or asks something.
  4. When it reports every task implemented and verified, start the reviewer the same way, in its own tagged pane, on the same worktree. Tell it the plan path, the implementer's pane, and: "You are the reviewer. Follow implement-plan's Review round and settle findings directly with the implementer. Don't edit files."
  5. When the review is settled, quit the reviewer and close its pane. Bring open disagreements to the user.
  6. When the implementer reports its PR, check the branch is fully pushed, quit the agent, close its pane, remove the worktree (see use-worktrees), and report the PR.

Otherwise, implement it yourself.

Implement

  1. Read the plan and its ADR. If something is unclear or looks wrong, ask before building on it.
  2. Work through the tasks in order. Build the simplest thing that meets each task, with no extra layers, options, or defensive code for cases that can't happen.
  3. Hand complex or parallel tasks to subagents.
  4. Add or update tests only where they protect behavior that matters (see awesome-tests).
  5. Verify every task yourself: run the checks the plan names, and check anything with a UI in a real browser with agent-browser. Set up any service or test data a check needs; don't leave testing to the user.
  6. Keep the plan file current. If you deviate, update the plan, or write a new ADR if the decision changed.
  7. When all tasks are done, get the review round. If the plan was handed to you, say you're ready and wait for the reviewer; otherwise start one yourself (its own herdr pane, or a subagent outside herdr).
  8. Commit, push, and open a pull request that says what changed, how you verified it, and what the review changed.
  9. Stop dev servers and close browsers. Leave the worktree to whoever created it.

Report the PR link, the verification, the review's outcome, and anything left open.

Review round

Every implementation gets one review by an agent that didn't write it, to catch what the author can't see, not to grow the code.

  • Reviewer: read the plan, the ADR, the full diff and the code it plugs into (see code-review). Run the checks and try to break it as a user would. Report only real, likely problems and removable complexity, each with a concrete failure scenario (or the complexity removed) and a fix. Skip niche edge cases, speculative hardening and style preferences.
  • Implementer: judge each finding instead of accepting it. Fix what is real with the simplest fix; reject, with a reason, fixes that add more complexity than the problem is worth.
  • Settle findings in direct discussion; unresolved ones go to the user. Review again only after substantial fixes.

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.