agentleFS
ngthluungthluu/my-dev-skills/skills/implement/SKILL.md

implement

Skill2 starsChanged 20 days agoMITView on GitHub

Implement a requested change or completed spec with test-driven development and independent code review.

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

25 rules. 0 are common, 25 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 171 edited versions we read keep these.

  • 25

    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:

    Use subagents only when useful, available, and permitted for bounded lookup, isolated auxiliary edits, extra checks, or read-only review.

    line 16
  2. Distinctive:

    Keep product decisions, architecture, and acceptance of results with the main agent.

    line 17
  3. Distinctive:

    Read the entire supplied spec and its required artifacts yourself.

    line 24
  4. Distinctive:

    Start from a spec's Suggested slices.

    line 34
  5. Distinctive:

    Follow this TDD cycle for behavioral changes, including delegated edits.

    line 41
  6. Distinctive:

    Write one behavior-focused test through the agreed public seam.

    line 43
  7. Distinctive:

    Run it and observe failure from missing behavior.

    line 44
  8. Distinctive:

    Run the focused test and neighboring tests green.

    line 46
  9. Distinctive:

    Write the least code that meets the contract.

    line 49
  10. Distinctive:

    Do not mirror the production algorithm, mock the behavior under test, or weaken a correct test to pass.

    line 56
  11. Distinctive:

    Avoid private-method tests, internal mocks, tautologies, conditional assertions, and checks that only assert no crash.

    line 57
  12. Distinctive:

    Add a contrasting or boundary case when one example permits constant output, a hard-coded special case, or ignored input.

    line 58
  13. Distinctive:

    Run focused tests and available typechecking during implementation.

    line 63
  14. Distinctive:

    Before review, read the whole diff.

    line 64
  15. Distinctive:

    After integration, run defined formatting, lint, typechecking, the full relevant test suite, and build.

    line 65
  16. Distinctive:

    When delegation is available and permitted, give two reviewers the contract, rules, baseline, diff, changed files, and verification …

    line 72
  17. Distinctive:

    Use reviewers who did not implement the changes.

    line 73
  18. Distinctive:

    If delegation is unavailable or prohibited, review the two axes separately yourself and disclose the limitation.

    line 74
  19. Distinctive:

    Requirements: check missing criteria, incorrect behavior, unsupported scope, and evidence at each public test seam.

    line 75
  20. Distinctive:

    Code: check correctness, maintainability, security, concurrency, errors, conventions, and test quality.

    line 76
  21. Distinctive:

    Ask what could be deleted or merged without changing behavior.

    line 77
  22. Distinctive:

    Check that tests catch plausible violations.

    line 78
  23. Distinctive:

    Verify each finding against the contract and code.

    line 92
  24. Distinctive:

    Ensure final verification covers all review fixes.

    line 100
  25. Distinctive:

    Do not commit, push, open a pull request, deploy, or publish unless requested.

    line 102

“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

Deliver the requested change or supplied spec completely. Follow repository instructions and preserve unrelated user changes.

Working rules

  • The main agent establishes the contract, plans behavior slices, implements core behavior, integrates changes, and verifies the result itself.
  • Use subagents only when useful, available, and permitted for bounded lookup, isolated auxiliary edits, extra checks, or read-only review.
  • Keep product decisions, architecture, and acceptance of results with the main agent. Handle small tasks directly.
  • Give agents the contract, relevant rules, baseline, evidence, and explicit file ownership. Reuse them with updated context.
  • Parallelize independent side tasks. Serialize writers unless files and dependencies are disjoint. Prohibit further delegation.
  • Require concise findings, changed files, commands, outcomes, and uncertainties. Verify results and keep raw reports out of chat.

Step 1: Establish the contract

  • Read the entire supplied spec and its required artifacts yourself. Otherwise use the current request as the contract.
  • Identify observable acceptance criteria and public test seams. Ask only about unresolved material product decisions.
  • Inspect repository rules, relevant code, prior decisions, and verification commands yourself. Confirm a spec's Where to change paths.
  • Resolve material unknowns through inspection and primary-source research. Check dates, retain links, and distinguish assumptions.
  • Reconcile evidence and record the starting Git ref, status, and relevant diff before mutations. Keep the agreed scope.

Step 2: Plan behavior slices

  • Decide boundaries, dependencies, and risks yourself. Use focused side investigations when useful and verify their evidence.
  • Divide work into small vertical slices, each delivering one observable behavior at an agreed public seam.
  • Start from a spec's Suggested slices. Change them only when the code requires it, and say why.
  • Name each slice's data shape first. Prefer one structure, such as a union, table, or state machine, over scattered conditionals.

Step 3: Implement with red and green tests

  • Implement core behavior slices yourself. Delegate only isolated auxiliary edits with clear criteria, public seams, and file ownership.
  • Auxiliary writers inspect the relevant source themselves before editing.
  • Follow this TDD cycle for behavioral changes, including delegated edits. Inspect red/green evidence and integrate before dependent work.
  1. Write one behavior-focused test through the agreed public seam.
  2. Run it and observe failure from missing behavior. Setup, fixture, and syntax errors do not count as red.
  3. Make the smallest implementation that passes, without anticipating later slices.
  4. Run the focused test and neighboring tests green. Refactor while tests stay green.
  5. Repeat for the next behavior instead of writing a batch of tests ahead of implementation.
  • Write the least code that meets the contract. Delete replaced paths, dead code, and redundant checks in touched code first.
  • Reuse existing helpers. Add no abstraction, option, or wrapper without a current second caller.
  • Validate external input once at its boundary. Do not recheck states that internal types already exclude.
  • Replace an internal API by migrating its callers and deleting it in the same change. Add no compatibility shim.
  • Comment only a non-obvious why that the code cannot show. No narration and no workaround justifications.
  • Assert outcomes through the real public path using expected values from requirements, known literals, or worked examples.
  • Do not mirror the production algorithm, mock the behavior under test, or weaken a correct test to pass.
  • Avoid private-method tests, internal mocks, tautologies, conditional assertions, and checks that only assert no crash.
  • Add a contrasting or boundary case when one example permits constant output, a hard-coded special case, or ignored input.
  • Repair tests that pass without the behavior. Observe a valid red state before continuing that slice.

Step 4: Verify the implementation

  • Run focused tests and available typechecking during implementation.
  • Before review, read the whole diff. Remove code that no criterion or test needs, then rerun affected checks.
  • After integration, run defined formatting, lint, typechecking, the full relevant test suite, and build. Resolve failures.
  • Also run a spec's Verification commands. Report missing or broken ones.
  • Extra checks may be side tasks. Run decisive acceptance checks yourself and inspect commands, outcomes, and failure evidence.

Step 5: Review independently

  • Compare changes with the recorded baseline. Investigate an unexpected empty diff.
  • When delegation is available and permitted, give two reviewers the contract, rules, baseline, diff, changed files, and verification evidence.
  • Use reviewers who did not implement the changes. Run read-only reviews in parallel, without further delegation.
  • If delegation is unavailable or prohibited, review the two axes separately yourself and disclose the limitation.
  • Requirements: check missing criteria, incorrect behavior, unsupported scope, and evidence at each public test seam.
  • Code: check correctness, maintainability, security, concurrency, errors, conventions, and test quality.
  • Ask what could be deleted or merged without changing behavior. Flag one-caller layers, speculative options, and guards hiding bugs.
  • Check that tests catch plausible violations. Flag mirrored algorithms, internal mocks, bypassed seams, and overfit fixtures.
  • Separate rule violations from judgment calls. Skip issues conclusively enforced by passing tools.
  • Review implementation changes, mentioning relevant pre-existing issues separately. Keep reports and unverifiable items explicit.
Axis: Requirements | Code
Verdict: Approve | Findings | Blocked
Findings:
- Critical | Important | Minor: file:line — violated requirement/rule; impact; concrete fix
Could not verify: ...

Step 6: Resolve review findings

  • Verify each finding against the contract and code. Fix every valid in-scope issue.
  • Handle core fixes yourself and auxiliary fixes with their original writer.
  • Rerun affected checks and give original reviewers the fix diff and evidence. Recheck open findings and new problems from fixes.
  • Repeat both full reviews when fixes materially change behavior or design.
  • Continue until both axes approve without material findings or a concrete external blocker prevents completion.

Step 7: Finish

  • Ensure final verification covers all review fixes. Reuse current passing evidence when no relevant change followed it.
  • Correct a supplied spec only for necessary factual corrections or approved decisions, preserving its intent.
  • Do not commit, push, open a pull request, deploy, or publish unless requested.
  • Report the outcome concisely:
Changed: <files and acceptance criteria>
Checks: <commands and outcomes, including red/green evidence>
Review: <requirements verdict; code verdict; fixes>
Limitations or blockers: <unverified items or required input>

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.