agentleFS
nelantonenelantone/cognivia/.agents/skills/cxp/SKILL.md

cxp

Skill1 starsChanged 3 months agoMITView on GitHub

Project-owned Codex Prompt orchestration workflow for generating or executing a structured repository task. Use only when the user explicitly invokes `$cxp`, writes `CXP`, or asks for the Cognivia CXP workflow; never select it implicitly.

The file

Show its rules and text

We split this skill into the rules it gives an agent

21 rules. About 0 are common, 13 are distinctive, and 8 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈13

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 8

    Very specific

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

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

    Use only when the user explicitly invokes $cxp, writes CXP, or asks for the Cognivia Codex Prompt workflow.

    line 17
  2. Distinctive:

    Do not select CXP implicitly for ordinary implementation, audit, refactor, documentation, commit-review, or handoff requests.

    line 22
  3. Distinctive:

    Every generated CXP prompt must begin with these five lines, in this order, with values filled in for the task:

    line 38
  4. Distinctive:

    After the header, include the objective, repository and branch, scope, protected areas, constraints, non-goals, expected files, acceptance …

    line 49
  5. Distinctive:

    If the user asked only for prompt generation, stop after producing the prompt.

    line 53
  6. Very specific:

    Use task-brief for request normalization, scope, constraints, non-goals, acceptance criteria, and validation contracts.

    line 58
  7. Very specific:

    Use architecture-audit for read-only repository inventory, dependencies, ownership, coupling, risks, and phases.

    line 60
  8. Very specific:

    Use safe-refactor for one approved behavior-preserving refactor phase.

    line 62
  9. Very specific:

    Use docs-update for evidence-backed maintained documentation changes.

    line 64
  10. Very specific:

    Use commit-review for non-mutating staged scope, commit grouping, messages, and readiness.

    line 66
  11. Very specific:

    Use session-handoff for factual continuity.

    line 68
  12. Distinctive:

    Read repository instructions and task-local guidance.

    line 74
  13. Very specific:

    Create or overwrite .cxp/CXP_HANDOFF.md, then return the concise terminal summary defined below.

    line 100
  14. Distinctive:

    Keep the handoff concise and factual.

    line 113
  15. Distinctive:

    commit readiness, blockers, split, exact paths, and proposed messages.

    line 122
  16. Distinctive:

    Follow AGENTS.md and every selected canonical skill.

    line 143
  17. Distinctive:

    Never reset, restore, revert, stash, discard, or delete existing work without exact authorization.

    line 146
  18. Distinctive:

    Never include secrets, credential values, private URLs, full provider responses, command transcripts, large logs, or full diffs in the …

    line 148
  19. Distinctive:

    Do not mask failures with cosmetic workarounds, delays, reloads, hidden overflow, broad exception handling, or silent continuation.

    line 150
  20. Distinctive:

    Use the validation ladder in AGENTS.md and the selected canonical skill.

    line 155
  21. Distinctive:

    Verify the handoff records current Git state, actual evidence, known issues, and one exact next action.

    line 164

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “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

CXP — Codex Prompt

Purpose

Generate or carry out a concise, repository-aware Codex prompt while preserving the CXP-specific scope, quality, validation, commit-readiness, and handoff contract. CXP orchestrates canonical skills; it does not own or duplicate their reusable procedures.

Trigger

Use only when the user explicitly invokes $cxp, writes CXP, or asks for the Cognivia Codex Prompt workflow.

Do not trigger

Do not select CXP implicitly for ordinary implementation, audit, refactor, documentation, commit-review, or handoff requests. Use the matching canonical responsibility skill directly.

Inputs

  • The requested outcome and whether CXP should generate a prompt, execute it, or do both.
  • Current repository instructions, Git state, authorization boundary, and unrelated work that must be preserved.
  • Scope, non-goals, affected boundaries, sensitive files, acceptance criteria, and focused validation.
  • Required browser, manual, provider, persistence, or deployment evidence.

Prompt contract

Every generated CXP prompt must begin with these five lines, in this order, with values filled in for the task:

ChatGPT/Codex:
recommended:
minimum:
reasoning recommended:
reasoning minimum:

After the header, include the objective, repository and branch, scope, protected areas, constraints, non-goals, expected files, acceptance criteria, validation, Git authorization, reporting contract, and exact next action. Keep the prompt focused enough to execute without turning a narrow request into a broad audit. If the user asked only for prompt generation, stop after producing the prompt; do not infer authorization to execute it.

Routing

  • Use task-brief for request normalization, scope, constraints, non-goals, acceptance criteria, and validation contracts.
  • Use architecture-audit for read-only repository inventory, dependencies, ownership, coupling, risks, and phases.
  • Use safe-refactor for one approved behavior-preserving refactor phase.
  • Use docs-update for evidence-backed maintained documentation changes.
  • Use commit-review for non-mutating staged scope, commit grouping, messages, and readiness.
  • Use session-handoff for factual continuity.
  • Apply only the skills required by the explicit request, in the order the work requires. The six canonical skills remain the responsibility owners.

Workflow

  1. Read repository instructions and task-local guidance. For expensive or agentic work, apply the repository provider guard before proceeding.
  2. Inspect the branch, staged, unstaged, and untracked state plus relevant diffs. Preserve unrelated work and identify sensitive paths without printing secret values.
  3. State the requested behavior, affected and unchanged boundaries, non-goals, expected files, risks, rollback point, and focused validation. Treat an uncertain cause as a hypothesis until repository or runtime evidence proves it.
  4. Generate the structured CXP prompt when requested, using the exact header above. Execute it only when the user also authorized implementation.
  5. Route each responsibility to its canonical skill. Make the smallest safe change and preserve behavior outside the approved scope.
  6. Apply CQM throughout: pragmatic clean code, short focused functions, clear UI/business-logic/RAG boundaries, focused tests, clear names, type hints where useful, explicit errors, and no unnecessary duplication.
  7. Validate according to the changed surface. Keep automated unit evidence, Streamlit AppTest evidence, real-browser evidence, manual checks, provider evidence, and documentation evidence distinct. Never claim that AppTest proves visual paint, DOM timing, or browser-only behavior.
  8. Explain the changes, inspect the complete relevant diff, and perform an honest final self-review. Report failures, skipped checks, unrelated changes, and unresolved assumptions without inflating completeness.
  9. Assess commit readiness without staging or committing automatically. State blockers, remaining manual checks, coherent commit boundaries, exact paths, sensitive files, and proposed messages when requested.
  10. Create or overwrite .cxp/CXP_HANDOFF.md, then return the concise terminal summary defined below.

Outputs

For prompt-generation requests, return the structured CXP prompt with the exact five-line header.

For executed or assessed work, create .cxp/CXP_HANDOFF.md even when the task is incomplete, blocked, or has failed validation. Treat explicit CXP invocation as authorization for this ignored compatibility artifact only. Do not create dated or task-specific CXP handoffs.

Keep the handoff concise and factual. Include:

  • request, date, and complete/incomplete/blocked status;
  • behavior and code changes;
  • architecture impact and technical debt;
  • files, areas, and sensitivity;
  • affected and unchanged boundaries;
  • sensitive-backend and risk review;
  • tests, compilation, lint, diff, browser/manual, and skipped checks;
  • commit readiness, blockers, split, exact paths, and proposed messages;
  • staged, unstaged, untracked, and committed Git state;
  • self-review result, unresolved assumptions, and one exact next action.

The final terminal response must be concise and include:

Status: complete / incomplete / blocked
Ready to commit: yes / no
Handoff: .cxp/CXP_HANDOFF.md

On macOS, optionally show this command for the user to run. Never execute it automatically:

pbcopy < .cxp/CXP_HANDOFF.md

Safety

  • Follow AGENTS.md and every selected canonical skill.
  • CXP grants no provider, dependency, product-edit, staging, commit, merge, push, deletion, or history-rewrite authority beyond the user's request.
  • Never reset, restore, revert, stash, discard, or delete existing work without exact authorization.
  • Never include secrets, credential values, private URLs, full provider responses, command transcripts, large logs, or full diffs in the handoff.
  • Do not mask failures with cosmetic workarounds, delays, reloads, hidden overflow, broad exception handling, or silent continuation.

Validation

  • Use the validation ladder in AGENTS.md and the selected canonical skill.
  • Require focused deterministic tests for changed behavior where applicable.
  • Distinguish checks actually run from recommended, unavailable, or skipped checks.
  • Give full completeness only when the cause or objective is verified, relevant checks pass, required browser/manual work is complete, sensitive changes are explained, and no known blocker remains.
  • Lower the self-review result when evidence is partial; never convert an inferred cause, unrun check, or unavailable browser into a passing claim.
  • Verify the handoff records current Git state, actual evidence, known issues, and one exact next action.

References

  • AGENTS.md — authoritative repository rules.
  • docs/agents/README.md — tooling ownership and supported workflow model.
  • docs/agents/SKILL_MIGRATION.md — CXP ownership and recovery record.

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.