agentleFS
scolladonscolladon/craft/skills/implementation/SKILL.md

implementation

Skill2 starsChanged 34 days agoMITView on GitHub

Craft phase 5 - execute the plan part by part, one part-implementer agent each, sequential, with per-part gates and one phase-boundary gate.

18 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use bug-fix, investigate and 3 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.
  • u9401066Skill

    bug-fix

    Structured bug fixing workflow with root cause analysis and regression prevention. Triggers: BF, 修 bug, fix bug, debug, 除錯, 錯誤, error, 問題, issue, 修復, fix, 故障, fault, 異常, exception, 失敗, failure, 不能用, broken, 壞掉.

    Also in 3 projects that use this

  • VercelSkill

    investigate

    Investigate an issue report, or analyze the constraints of a proposed change, with strict, evidence-driven diagnosis.

    Also in 3 projects that use this

  • MicrosoftSkill

    playwright-cli

    Automate browser interactions, test web pages and work with Playwright tests.

    Also in 4 projects that use this

  • MicrosoftSkill

    release

    Use ONLY when the user explicitly asks to cut/publish a Huabu desktop release or tag a version. Performs an outward action — bumps the version in apps/desktop/package.json, commits + annotated-tags vX.Y.Z, and pushes the branch + tag. That tag push auto-builds the macOS/Windows installers and publishes a GitHub Release. Provides `status`, `bump`, `tag-push`, and a one-command `run` (bump → tag-push).

    Also in 4 projects that use this

  • SentrySkill

    skill-creator

    Create new agent skills following the Agent Skills specification. Use when asked to "create a skill", "add a new skill", "write a skill", "make a skill", "build a skill", or scaffold a new skill with SKILL.md. Guides through requirements, writing, registration, and verification.

    Also in 4 projects that use this

The file

We split this skill into the rules it gives an agent

1 rule. 0 are common, 0 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 17 edited versions we read keep these.

  • 0

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

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

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

    Phase-boundary gate: after the LAST part, run gates.phase once in-session.

    line 63

“Common” is measured: kept by at least 6 in 10 of the 17 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

craft:implementation

Preamble (always runs — non-overridable)

  1. Manifest read (lint if standalone). Probe gates: gates.part (else: repo test runner over touched files; else fall back to gates.phase per part — slower, never gateless) and gates.phase (else: repo validate/check/test script; if nothing exists, REFUSE to run this phase — a workflow without any gate is not this workflow).
  2. The plan must exist and pass plan-lint.sh (re-run it; cheap) — unless no plan exists at <plan-dir>/<slug>.md and the run record holds a NO-OP(planning): line (grep -F 'NO-OP(planning):' on the ledger path that run-ledger.sh locate or the reorient block yields). Then the phase is a no-op: spawn nothing, append no PART line, skip Procedure 1–2, and run gates.phase once in-session (the Procedure 3 gate — the phase is never gateless, even here; a manifest override: cannot skip it). On green, one run-ledger.sh append <run-id> implementation call carries NO-OP(implementation): no plan parts — planning recorded NO-OP(planning), GATE(implementation): green and PHASE-DONE(implementation): no-op — no plan parts. On red, append the NO-OP line and GATE(implementation): red without PHASE-DONE, fix and commit as Procedure 3 states, re-run the gate, and append its result: PHASE-DONE lands only with a green gate, so a rebuild never reads a red phase as closed. A plan file that exists next to a NO-OP(planning): line is a contradiction: escalate { implementation, reason, ≤3 options }, never guess which wins. No plan and no NO-OP(planning): line (standalone included): refuse, as before.
  3. Memory read/write surface (advisory). READS: gate-cmd entry for this repo as an advisory hint — if a gate command was previously recorded, skip re-discovery but still run it (the gate is sacred; the hint only saves the probe, never the execution). A miss falls through to the full gate probe. WRITES (appended to the run record as produced; saved to the store once at Done): the gate/test command discovered this run — stored as the BARE command only, with any leading env/secret assignment prefix stripped (never TOKEN=… npm test or a command carrying a credential), since the store is committed; per-part size + pass/blocked outcome for each implemented part.

Procedure (default body — a manifest override: replaces everything below)

  1. Execute parts top-to-bottom, one craft:part-implementer per part, sequential — never two agents writing one tree concurrently. Each spawn carries: the working directory; the plan path (the agent reads the part there) + the part's ### Context block verbatim AND any load-bearing deltas — hand the pre-chew over so the agent never re-greps the codebase, but do NOT re-transcribe the whole committed part; the design doc path; the resolved part gate; the commit message from the plan; global + implementation-phase context: files verbatim.
  2. After each agent returns, verify before launching the next: the commit exists and matches the part promise (git log, git show --stat). Once verified, in that same verification's Bash call or the very next one, append PART(<n>): <sha> size=<size> outcome=<pass|blocked> via "${CRAFT_ROOT:-${CLAUDE_PLUGIN_ROOT}}/scripts/run-ledger.sh" append <run-id> implementation — <size> is one kebab-case shape label for the part (e.g. pure-module, docs-prose), ? when unknown; a label with a space breaks the token. A landed commit found without a PART line on resume is verified first, then gets its line. Spot-check conventions on the diff. Every IMPACT( line and deferred observation in the handback gets its verdict under the run skill invariant Impacts: fold or defer at discovery before the next part spawns. Failed/blocked part → fix in-session or escalate with the agent's options; dead agent → fresh respawn from the plan part (artifact handoff); never relaunch blindly.
  3. Phase-boundary gate: after the LAST part, run gates.phase once in-session. It MUST be green before the review phase — fix and commit anything it surfaces (fix(<scope>): close gate gap after part N).

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.