implement-spec
Skill3 starsChanged 4 months agoApache-2.0View on GitHub
Implement a spec from ai/lab/specs/ exactly as written — zero design decisions, stop-and-report on any spec-vs-reality conflict, tests are the definition of done. Built for lighter implementation models.
129 other projects use a version of this.
Adopted by
Most recently taken up by
- halilural/imprimaturtheir own version, first seen Nov 16, 2026
- dorian-akhs/skillstheir own version, first seen Nov 16, 2026
- clnnn/playbooktheir own version, first seen Nov 16, 2026
- ericmasiello/skillstheir own version, first seen Nov 16, 2026
- zanellig/skillstheir own version, first seen Nov 16, 2026
- jjsizemore/jjskillstheir own version, first seen Nov 16, 2026
Used with
Projects with it also use implement-spec-with-codex, close-spec and 9 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.
dzhngSkillimplement-spec-with-codex
Run [implement-spec](../implement-spec/SKILL.md) with Codex doing the implementation passes while you orchestrate, integrate, and review. Use when the user asks to implement a spec with Codex, or to run implement-spec with Codex as the implementer.
Also in 4 projects that use this
dzhngSkillclose-spec
When a spec under specs/<feature>/ is done shipping (a write-spec build, or any planned task that produced a spec), archive it to specs/done/ and rewrite it from a build-plan into a durable rationale — the why, the principles, the invariants — pointing back to the real code for the how. Use when implementation has landed and the plan no longer matches what shipped, or the user says a feature/spec is finished. Pairs with [write-spec](../write-spec/SKILL.md) (the plan this closes) and [review](../review/SKILL.md) (run before closing).
Also in 4 projects that use this
wolverin0Skillaudit-decisions
Standardize remediation decisions, baseline-policy entries, and legal-review banners for accepted, deferred, or policy-blocked audit findings.
Also in 4 projects that use this
saffron-healthSkillget-pr-reviews
Get nicely formatted output of the last review comments on a PR. Used internally by other skills.
Also in 3 projects that use this
saffron-healthSkillpush-everything
Stage ALL changes in the repository (not just session changes), commit, and push. Use when you need to commit all pending changes.
Also in 3 projects that use this
transitive-bullshitSkillcompare-screenshots
Compare screenshots against the intended design, distinguishing approved references from historical baselines. Use for iterative implementation matching, before/after visual review, objective image telemetry, or checking a lone capture for flat, empty, or badly framed content.
Also in 4 projects that use this
- +4 more from dzhng
Other versions129 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 129
- mattpocock/skills★ 325ksame name
- formbricks/formbricks★ 13ksame name
- ArcReel/ArcReel★ 6.6ksame name
- vinvcn/mattpocock-skills-zh-CN★ 5.8ksame name
- withcoral/coral★ 5ksame name
- dzhng/jevgrep★ 3.2ksame name
- dzhng/skills★ 1.5ksame name
- a16z/jolt★ 1.1ksame name
- cashew-labs/libretto★ 911same name
- saffron-health/libretto★ 911same name
119 more not listed
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.
In this repository32 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- .cursor/rules/add-feature.mdcCursor rule
- .cursor/rules/adversarial-audit.mdcCursor rule
- .cursor/rules/ai-knowledge-layer.mdcCursor rule
- .cursor/rules/check-drift.mdcCursor rule
- .cursor/rules/cold-start.mdcCursor rule
- .cursor/rules/create-feature-catalog.mdcCursor rule
- .cursor/rules/fix-bug.mdcCursor rule
- .cursor/rules/perform-feature-add-simulation.mdcCursor rule
- .cursor/rules/post-cold-start-verification.mdcCursor rule
- .cursor/rules/review-agent-config.mdcCursor rule
- .cursor/rules/review-change.mdcCursor rule
- .cursor/rules/verify-ai-readiness.mdcCursor rule
- add-featureSkill
- cold-startSkill
- deep-testSkill
- fix-bugSkill
- review-changeSkill
- add-featureSkill
- adversarial-auditSkill
- check-docsSkill
- check-driftSkill
- cold-startSkill
- create-feature-catalogSkill
- fix-bugSkill
- perform-feature-add-simulationSkill
- post-cold-start-verificationSkill
- review-agent-configSkill
- review-changeSkill
- verify-ai-readinessSkill
The file
We split this skill into the rules it gives an agent
11 rules. 0 are common, 6 are distinctive, and 5 are very specific to this project.
0
Common
Most of the 121 edited versions we read keep these.
6
Distinctive
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
- Distinctive: line 19
Read the ENTIRE spec file, top to bottom, before writing any code.
- Distinctive: line 20
Read any rationale/lessons-learnt doc the spec's preamble names — once.
- Very specific: line 21
Read ai/guide/MODULE_MAP.md (Stability column) and ai/guide/CONVENTIONS.md.
- Distinctive: line 22
Do NOT crawl the tree. Open other files only when the spec's touch list or your current step requires them.
- Very specific: line 28
Every file in the spec's touch list marked "modify" exists.
- Distinctive: line 34
If at ANY point the spec conflicts with what you find in the code, the docs, or a failing behaviour — the function signature differs, a …
- Very specific: line 40
Do NOT pick the "obvious" interpretation.
- Distinctive: line 46
Follow the spec's own ordering and any code skeleton it provides — structure, export names, and file layout are contractual, not …
- Distinctive: line 60
If the spec has acceptance criteria, walk the list and mark each one explicitly: met / not met / blocked-on-question.
- Very specific: line 73
NEVER flip [inferred] → [verified], and never edit an existing [verified] signature.
- Very specific: line 77
If the spec implements a ai/lab/ROADMAP.md Planned feature, move its row from Planned to Shipped, cross-linking the spec, the WORKLOG row, …
“Common” is measured: kept by at least 6 in 10 of the 121 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.
Files
<!-- Copyright (c) 2026 Kunal Suri (CEA LIST). All rights reserved. -->Implement the spec at: $ARGUMENTS (If no path was given, ask for one and stop. Implement ONLY this spec.)
You are the implementer, not the designer. Every design decision has already been made in the spec. Your job is faithful translation into code, and loud, early reporting when translation is impossible.
Protocol (follow in order, no skipping)
0. Load context — exactly this, no more
- Read the ENTIRE spec file, top to bottom, before writing any code.
- Read any rationale/lessons-learnt doc the spec's preamble names — once.
- Read
ai/guide/MODULE_MAP.md(Stability column) andai/guide/CONVENTIONS.md. - Do NOT crawl the tree. Open other files only when the spec's touch list or your current step requires them.
1. Pre-flight (all must pass before you write code)
npm testpasses on the clean checkout (record the pass count).node install.mjs verify . --strictpasses.- Every file in the spec's touch list marked "modify" exists; confirm none is
frozenor?in MODULE_MAP.md. If one is → STOP AND REPORT. - Restate to yourself the spec's hard-constraints table (if present). Those constraints outrank your habits and your training priors.
2. THE PRIME RULE — stop-and-report, never improvise
If at ANY point the spec conflicts with what you find in the code, the docs, or a failing behaviour — the function signature differs, a file isn't where the spec says, an example contradicts a rule, a specified test cannot pass as written — you MUST stop and report:
- the spec section, the file:line evidence, and the smallest question whose
answer unblocks you.
Do NOT pick the "obvious" interpretation. Do NOT patch around it. Do NOT widen
scope to fix it. A wrong guess here costs more than the pause. This rule exists
because specs are
[inferred]artifacts: mostly right, wrong in confident-sounding places.
3. Implement
- Follow the spec's own ordering and any code skeleton it provides — structure, export names, and file layout are contractual, not suggestions.
- Touch ONLY files in the spec's touch list. No new dependencies unless the spec adds them. No drive-by refactors, renames, or "improvements".
- License header on every new file, copied from a neighbouring file.
- Match surrounding code style (imports, naming, error handling) over generic best practice.
4. Verify — the tests are the definition of done
- Implement the spec's ENTIRE test plan (every numbered test), in the harness style the spec names. A test you cannot implement as specified is a §2 stop-and-report, not a skip.
npm testgreen (old count + new tests),node install.mjs verify . --strictclean. Paste real output; never summarize a test run you didn't execute.- If the spec has acceptance criteria, walk the list and mark each one explicitly: met / not met / blocked-on-question.
5. Review gate (the engineering loop's Review step — not optional)
- Request
/review-changeon the completed diff in a FRESH context (a session that did not write the code), perai/lab/specs/SPEC_engineering-loop.md. You implement; you do not approve your own work. - The review lands in
ai/lab/reviews/and its verdict is input to the human's merge decision — implementation "done" without a review is not loop-complete.
6. Knowledge updates (part of the change, not an afterthought)
- Execute the spec's "knowledge update on completion" checklist. All new/changed
rows and entries are tagged
[inferred]. - NEVER flip
[inferred]→[verified], and never edit an existing[verified]signature. That is the human's move alone. - Append the WORKLOG entry linking spec, review, commits, and test evidence
(verify scans these links — broken ones fail
--strict). - If the spec implements a
ai/lab/ROADMAP.mdPlanned feature, move its row from Planned to Shipped, cross-linking the spec, the WORKLOG row, and the PR.
7. Final report (in this order)
- Acceptance-criteria scorecard.
- Verbatim tail of
npm testandverify --strictoutput. - Deviations from the spec: list each one with its §2 report — the expected count is ZERO; anything else should already have stopped you.
- Files changed vs. the spec's touch list (must match exactly).
- Review status: link to the
/review-changerequest/result for the human's merge decision.
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
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.

