2Tricky4u

coverage

Completeness gate - judge each `## Deliverables` checklist item in the plan against the actually-built tree (present? substantive, not a stub? meets its acceptance criterion?) and file one finding per gap. Read-only; decides whether a feature is genuinely done, not just tests-green.

Code & Development Listed

debug

Systematic root-cause debugging. Use when a bug, test failure, crash, or unexplained behavior needs fixing - especially when the cause is not yet understood. Four phases; fixing before understanding is forbidden.

Code & Development Listed

dual-review

Independent two-model review of the current diff - Claude's code-reviewer subagent and OpenAI Codex review the same changes without seeing each other's findings, then results are merged and only confirmed findings are fixed. Use before committing significant changes.

plan-review

Cross-model adversarial review of an implementation plan before executing it. Use after plan mode produces a plan, or when the user asks to critique/review/stress-test a plan. OpenAI Codex critiques the plan, debate is bounded to 2 rounds, output is a revised plan plus unresolved disagreements for the human.

spec

Apply ONE scoped change to a ritual feature document (its spec.md or plan.md) from a chat message. Use when ritual's spec-chat invokes you to author or refine a document - you edit the file in place and report the change in one line. Single-shot and non-interactive by design.

tdd

Test-first implementation loop with cross-model test design. Use when starting implementation of an approved plan or a nontrivial feature - OpenAI Codex designs the test list independently from the spec, Claude writes failing tests first, then implements until green.

brainstorm

Socratic discovery BEFORE planning - use when an idea or feature is fuzzy, when the user says "I want something like…", or before writing a spec. Converges on a written spec; explicitly the step before plan mode and /plan-review.

changelog

Generate Keep-a-Changelog release notes from a git range. Use for "write the changelog", "release notes for vX", or before tagging a release.

Code & Development Listed

commit

Write a Conventional Commit from the actual staged diff and commit it. Use when the user asks to commit, or when work is done and needs committing.

consensus

Third-model arbitration of ONE genuinely contested finding or design question via pal-mcp-server's stance-steered consensus (needs the pal MCP server + a Gemini key). Use ONLY when cross-model review left a critical/major disagreement unresolved - the evidence says ungrounded debate is the weakest pattern, so this is a last resort, not a routine step.

deps-audit

Dependency security + license audit for the current project (supply-chain complement to /security-review, which covers the code itself). Use for "audit dependencies", "any vulnerable packages?", "license check", or before a release.

docs

Generate or refresh project documentation (README sections, module docs, usage guides) from the actual code. Use for "document this", "update the README", "write usage docs".

Data & Documents Listed

document

Produce a deliverable document file (.docx or .pdf) from content - reports, specs, summaries - via markdown + pandoc. Use when the user needs a shareable file, not project docs.