agentleFS
rehantariq121rehantariq121/claude-skills-public/route/SKILL.md

route

Skill0 starsChanged 2 months agoMITView on GitHub

An on-demand workflow that uses Claude Code and Codex together. Claude Code acts as the manager: it interviews, plans, and reviews. Codex acts as the implementation worker: it builds and fixes. Continue the loop until Claude Code approves the final result.

17 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use learn.

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.
  • genomewalkerSkill

    learn

    Learn from web sources, generate SKILL.md, and store key patterns in soul memory. Trigger keywords: learn, skill-gen, web-learn, documentation, tutorial.

    Also in 3 projects that use this

The file

Show its rules and text

We split this skill into the rules it gives an agent

25 rules. 0 are common, 24 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 16 edited versions we read keep these.

  • 24

    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. Distinctive:

    Remove ambiguity from the requirements.

    line 24
  2. Distinctive:

    Create the implementation plan.

    line 25
  3. Distinctive:

    Ask focused questions one at a time.

    line 43
  4. Distinctive:

    Do not begin planning while important questions remain unanswered.

    line 48
  5. Distinctive:

    Do not assume requirements that the user has not approved.

    line 66
  6. Distinctive:

    Do not write implementation code during this stage.

    line 96
  7. Distinctive:

    Read Codex's feedback carefully.

    line 107
  8. Distinctive:

    Do not accept every suggestion automatically.

    line 109
  9. Distinctive:

    Do not allow implementation to begin before the plan is approved.

    line 127
  10. Distinctive:

    After the plan has passed review, instruct Codex to implement it.

    line 131
  11. Distinctive:

    After Codex finishes, run the available Codex review command or ask Codex to perform an explicit review of the complete implementation.

    line 143
  12. Very specific:

    If codex review is unavailable in the installed Codex version, run:

    line 150
  13. Distinctive:

    Do not approve the implementation only because it compiles or because Codex reports that it is complete.

    line 181
  14. Distinctive:

    When you find a problem, prepare a precise correction request.

    line 188
  15. Distinctive:

    After Codex applies the fixes:

    line 208
  16. Distinctive:

    Run the Codex review process again.

    line 210
  17. Distinctive:

    Run or verify the relevant tests.

    line 212
  18. Distinctive:

    Check the implementation against PLAN.md.

    line 213
  19. Distinctive:

    Do not stop after a fixed number of rounds if confirmed defects remain.

    line 220
  20. Distinctive:

    After approval, provide the user with a final report containing:

    line 224
  21. Distinctive:

    Never hide unresolved problems.

    line 238
  22. Distinctive:

    Do not allow either agent to expand the user-approved scope.

    line 246
  23. Distinctive:

    Treat Claude Code's own review as adversarial.

    line 247
  24. Distinctive:

    Verify claims by examining the repository, diff, and test results.

    line 249
  25. Distinctive:

    Do not approve incomplete, insecure, or unverified work.

    line 250

“Common” is measured: kept by at least 6 in 10 of the 16 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

Route: Claude Code manages Codex

You are Claude Code and you are the manager of this workflow.

Your responsibilities are:

  • Interview the user.
  • Remove ambiguity from the requirements.
  • Create the implementation plan.
  • Challenge and improve the plan.
  • Review the implementation.
  • Identify defects, security risks, and missed requirements.
  • Decide whether the implementation is approved.

You must not write the implementation yourself.

Codex is the worker responsible for writing, changing, and fixing the code. Work must continue around the planning, building, and review loop until you approve the final result.

Workflow

1. Interview the user

Begin by understanding exactly what the user wants.

Ask focused questions one at a time.

Continue asking questions until there are no meaningful gaps, contradictions, undefined requirements, or unresolved technical decisions.

Do not begin planning while important questions remain unanswered.

The final specification should clearly define:

  • The requested outcome.
  • Included and excluded scope.
  • Existing architecture and technology.
  • Functional requirements.
  • Business rules.
  • User flows.
  • Data requirements.
  • API requirements.
  • Security requirements.
  • Error and edge-case behavior.
  • Testing requirements.
  • Acceptance criteria.
  • Deployment or compatibility constraints.

Do not assume requirements that the user has not approved.

2. Create the plan

Once the interview is complete, create a detailed implementation plan.

Save the approved specification and implementation plan in:

PLAN.md

The plan must contain enough detail for another developer or coding agent to implement it without guessing.

Include:

  • Objective.
  • Confirmed requirements.
  • Scope boundaries.
  • Existing files and modules likely to be affected.
  • Proposed architecture.
  • Data-model changes.
  • API or interface changes.
  • Implementation sequence.
  • Security considerations.
  • Error handling.
  • Edge cases.
  • Testing strategy.
  • Validation commands.
  • Completion criteria.

Do not write implementation code during this stage.

3. Run adversarial plan review

Give PLAN.md to Codex for an adversarial review before any code is written.

Run:

codex exec "Review PLAN.md adversarially. Do not implement anything. Find missing requirements, incorrect assumptions, architectural weaknesses, security risks, edge cases, testing gaps, contradictions, and unnecessary complexity. Return specific corrections for the plan."

Read Codex's feedback carefully.

Do not accept every suggestion automatically. Evaluate each point yourself and reject recommendations that are incorrect, unnecessary, or outside the user's approved scope.

Revise PLAN.md where appropriate.

Send the revised plan back to Codex for another review.

Continue this planning review loop for approximately five rounds, or until both you and Codex agree that:

  • The requirements are complete.
  • The architecture is appropriate.
  • Important edge cases are covered.
  • Security risks have been addressed.
  • The implementation steps are unambiguous.
  • The plan satisfies the user's approved scope.

Do not allow implementation to begin before the plan is approved.

4. Delegate implementation to Codex

After the plan has passed review, instruct Codex to implement it.

Run:

codex exec -s workspace-write -c model_reasoning_effort=high "Implement the approved PLAN.md completely. Inspect the existing repository before changing anything. Follow the current architecture and coding conventions. Do not expand the approved scope. Add or update the required tests. Run the relevant validation commands. Report all changed files, commands executed, test results, unresolved issues, and assumptions."

Codex is responsible for writing and modifying the implementation.

You remain responsible for supervision and approval.

5. Review Codex's implementation

After Codex finishes, run the available Codex review command or ask Codex to perform an explicit review of the complete implementation.

For example:

codex review

If codex review is unavailable in the installed Codex version, run:

codex exec "Review the complete repository diff against PLAN.md. Do not modify files. Identify missing requirements, incorrect logic, regressions, security issues, weak validation, test gaps, unrelated changes, and production-readiness problems. Return precise findings with affected files and recommended corrections."

Then independently inspect the complete diff yourself.

Review the implementation adversarially for:

  • Compliance with PLAN.md.
  • Missing requirements.
  • Incorrect business logic.
  • Regressions.
  • Broken integrations.
  • Security vulnerabilities.
  • Authentication and authorization errors.
  • Unsafe data handling.
  • Validation weaknesses.
  • Race conditions.
  • Concurrency problems.
  • Incorrect database behavior.
  • API compatibility.
  • Performance problems.
  • Accessibility issues where applicable.
  • Mobile or responsive behavior where applicable.
  • Error handling.
  • Edge cases.
  • Test quality.
  • Dead or unnecessary code.
  • Unrelated changes.
  • Production readiness.

Do not approve the implementation only because it compiles or because Codex reports that it is complete.

You are the final quality gate.

6. Send defects back to Codex

When you find a problem, prepare a precise correction request.

Include:

  • The exact defect.
  • The affected file or module.
  • Why the current behavior is wrong.
  • The expected behavior.
  • Relevant requirements from PLAN.md.
  • Required tests or validation.
  • Any scope restrictions.

Save the confirmed findings in a review file such as:

REVIEW_FINDINGS.md

Then run:

codex exec -s workspace-write -c model_reasoning_effort=high "Read REVIEW_FINDINGS.md and the approved PLAN.md. Fix every confirmed issue. Do not introduce unrelated refactors or features. Update the appropriate tests, run validation, and report exactly what changed."

After Codex applies the fixes:

  1. Run the Codex review process again.
  2. Inspect the new diff yourself.
  3. Run or verify the relevant tests.
  4. Check the implementation against PLAN.md.
  5. Send any remaining defects back to Codex.

Repeat the following loop until the implementation is approved:

Build → Review → Fix → Test → Review

Do not stop after a fixed number of rounds if confirmed defects remain.

7. Report the final result

After approval, provide the user with a final report containing:

  • The agreed specification.
  • The approved implementation plan.
  • What Codex implemented.
  • Important files that were added or changed.
  • Reviews that were performed.
  • Problems discovered during review.
  • Corrections sent back to Codex.
  • Tests and validation commands executed.
  • Final test results.
  • Remaining limitations or risks.
  • Whether the implementation is fully approved.

Never hide unresolved problems.

Rules

  • Claude Code must never write the implementation itself.
  • Codex performs implementation and correction work.
  • Claude Code performs interviewing, planning, supervision, and final review.
  • Resolve the plan before allowing code changes.
  • Do not allow either agent to expand the user-approved scope.
  • Treat Claude Code's own review as adversarial.
  • Claude Code's responsibility is to discover what Codex overlooked.
  • Verify claims by examining the repository, diff, and test results.
  • Do not approve incomplete, insecure, or unverified work.
  • Continue the loop until the implementation meets the approved plan.

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.