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.
In this repository9 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- architecture-auditSkill
- commit-reviewSkill
- docs-updateSkill
- safe-refactorSkill
- session-handoffSkill
- task-briefSkill
- cxpSkill
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
- Very specific: line 17
Use only when the user explicitly invokes $cxp, writes CXP, or asks for the Cognivia Codex Prompt workflow.
- Distinctive: line 22
Do not select CXP implicitly for ordinary implementation, audit, refactor, documentation, commit-review, or handoff requests.
- Distinctive: line 38
Every generated CXP prompt must begin with these five lines, in this order, with values filled in for the task:
- Distinctive: line 49
After the header, include the objective, repository and branch, scope, protected areas, constraints, non-goals, expected files, acceptance …
- Distinctive: line 53
If the user asked only for prompt generation, stop after producing the prompt.
- Very specific: line 58
Use task-brief for request normalization, scope, constraints, non-goals, acceptance criteria, and validation contracts.
- Very specific: line 60
Use architecture-audit for read-only repository inventory, dependencies, ownership, coupling, risks, and phases.
- Very specific: line 62
Use safe-refactor for one approved behavior-preserving refactor phase.
- Very specific: line 64
Use docs-update for evidence-backed maintained documentation changes.
- Very specific: line 66
Use commit-review for non-mutating staged scope, commit grouping, messages, and readiness.
- Very specific: line 68
Use session-handoff for factual continuity.
- Distinctive: line 74
Read repository instructions and task-local guidance.
- Very specific: line 100
Create or overwrite .cxp/CXP_HANDOFF.md, then return the concise terminal summary defined below.
- Distinctive: line 113
Keep the handoff concise and factual.
- Distinctive: line 122
commit readiness, blockers, split, exact paths, and proposed messages.
- Distinctive: line 143
Follow AGENTS.md and every selected canonical skill.
- Distinctive: line 146
Never reset, restore, revert, stash, discard, or delete existing work without exact authorization.
- Distinctive: line 148
Never include secrets, credential values, private URLs, full provider responses, command transcripts, large logs, or full diffs in the …
- Distinctive: line 150
Do not mask failures with cosmetic workarounds, delays, reloads, hidden overflow, broad exception handling, or silent continuation.
- Distinctive: line 155
Use the validation ladder in AGENTS.md and the selected canonical skill.
- Distinctive: line 164
Verify the handoff records current Git state, actual evidence, known issues, and one exact next action.
≈ “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.
Files
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-brieffor request normalization, scope, constraints, non-goals, acceptance criteria, and validation contracts. - Use
architecture-auditfor read-only repository inventory, dependencies, ownership, coupling, risks, and phases. - Use
safe-refactorfor one approved behavior-preserving refactor phase. - Use
docs-updatefor evidence-backed maintained documentation changes. - Use
commit-reviewfor non-mutating staged scope, commit grouping, messages, and readiness. - Use
session-handofffor 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
- Read repository instructions and task-local guidance. For expensive or agentic work, apply the repository provider guard before proceeding.
- Inspect the branch, staged, unstaged, and untracked state plus relevant diffs. Preserve unrelated work and identify sensitive paths without printing secret values.
- 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.
- Generate the structured CXP prompt when requested, using the exact header above. Execute it only when the user also authorized implementation.
- Route each responsibility to its canonical skill. Make the smallest safe change and preserve behavior outside the approved scope.
- 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.
- 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.
- 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.
- 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.
- 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.mdOn macOS, optionally show this command for the user to run. Never execute it automatically:
pbcopy < .cxp/CXP_HANDOFF.mdSafety
- Follow
AGENTS.mdand 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.mdand 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.
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.

