Claude Skills by xiongxianfei

- Status: done - Owner: maintainers - Start date: 2026-04-29 - Last updated: 2026-04-30 - Related issue or PR: PR #23 - Supersedes: none - broad_smoke_required: true - broad_smoke_reason: Planned initiative touches governance, public README ownership, canonical skills, generated Codex skill mirrors, generated public adapter packages, lifecycle-managed artifacts, and proposal workflow guidance.

RigorLoop has a clear workflow contract, a public README, and a growing set of lifecycle skills, but it does not yet have a compact canonical vision document that answers what the project is, who it is for, what it refuses to become, and how future proposals should be checked for fit.

The `learn` skill is now correctly classified as periodic or explicitly invoked, but its final artifact model is still deferred. The current guidance says scheduled follow-ups and no-learn rationales use temporary contributor-visible or review-visible surfaces until a focused learn refactor defines the final model.

Create or update technical architecture after Proposal Review and before the specification is finalized.

Diagnose or fix unexpected behavior, failing evidence, incidents, regressions, and bug reports with bounded authority and identity-stable proof.

Author or review repository-owned CI workflows, validation automation, and related configuration when coverage, speed, caching, permissions, triggers, or ordinary hardening needs maintenance.

Review an implementation slice against the actual diff, approved design and delivery packages, tests, and validation evidence, then record findings or a clean first-pass review. Use after implement hands off a milestone or when implementation review is requested. Use design-review, delivery-review, verify, or pr for those gates instead.

Create or update the repository's governing AI-development principles. Use when starting a project, adopting this skill pack, onboarding a brownfield codebase, or when recurring mistakes show that project rules are unclear.

Independently review one plan-centered delivery package before implementation. Judge safe sequencing and verification adequacy together; use plan for delivery and verification allocation and spec for behavioral gaps.

Independently review architecture, specification, applicable ADRs, and accepted proposal constraints as one design package before delivery planning. Use design-review for design coherence and package authority; use architecture and spec for authorship.

Expand and challenge an idea before proposal, specification, or planning. Use at the early ideation stage when the problem, user value, scope, architecture direction, or product strategy is not settled.

Implement one approved milestone or isolated implementation request with tests or proof first, then hand it to code-review with validation evidence. Use when requirements, scope, and validation commands are clear enough to code. Use bugfix for defect reproduction/fix loops, code-review to review implementation, verify for final readiness, and pr for PR handoff.

Capture durable lessons after implementation, review, verification, or incidents. Use when recurring mistakes, systemic gaps, or explicit retrospective work should guide future contributors.

Create or revise a stable execution plan after proposal, spec, and architecture are settled enough to implement. Use for multi-file, multi-component, risky, migration-heavy, or milestone-based work that needs reviewable implementation slices, verification allocation, validation commands, recovery paths, and dependencies. Do not use plan to choose product direction, write code, review diffs, update workflow routing or existing planned work, verify branch readiness, or open PRs.

Prepare a completed, verified change for pull request review. Use when the branch is ready or nearly ready and the agent should summarize the real diff, validation evidence, spec compliance, risks, and reviewer notes.

Build, refresh, or audit a current-state repository orientation map. Use when architecture, module boundaries, runtime flow, data flow, test layout, CI, ownership, or change placement is unclear. This skill records observed repository structure and bounded inferences; it does not invent future architecture or act as a backlog.

Review a concise proposal and its embedded feasibility evaluation before Design. Use when the user asks whether a direction is valuable, bounded, feasible, vision-aligned, and specific enough to pursue without prematurely settling downstream decisions. Use proposal to write proposals; use design-review, delivery-review, code-review, verify, or pr for later gates.

Create a concise direction-approval proposal before Design. Use when the user has a selected direction, explored options, problem statement, or issue that needs its challenge, goals, bounds, governing principle, direction, and feasibility recorded. Use proposal-review to review an existing proposal; use architecture and spec for detailed Design decisions.

Validate assumptions before proposal, spec, architecture, or implementation. Use when the workflow depends on uncertain technical behavior, domain facts, user expectations, dependencies, platform limits, standards, laws, pricing, performance, or operational constraints.

Author observable feature contracts before delivery planning or implementation; Design Review approves the specification with architecture.

Run final verification and produce durable change rationale only on success before PR handoff. Use to verify artifact-code-test coherence, requirement coverage, validation commands, CI readiness, drift, release safety, and scoped direct validation checks.

Produce or update the project vision and matching README front-matter at project genesis or when current vision no longer reflects the project. This skill is upstream of the per-change workflow.

Orchestrate the full spec-driven, test-driven agentic development lifecycle. Use when starting, resuming, auditing, or routing work through the standard RigorLoop workflow. This skill assesses workflow state, enforces artifact order, and keeps exploration, specification, architecture, planning, tests, implementation, review, rationale, verification, PR, and learning connected.

- [Rename `ci` to `ci-maintenance` and Add Risk-Scoped GitHub Workflow Authoring Support](../docs/proposals/2026-05-26-ci-maintenance-skill-rename-and-workflow-authoring.md), accepted. - Proposal review records: - [proposal-review-r1](../docs/changes/2026-05-26-ci-maintenance-skill-rename-and-workflow-authoring/reviews/proposal-review-r1.md) - [proposal-review-r2](../docs/changes/2026-05-26-ci-maintenance-skill-rename-and-workflow-authoring/reviews/proposal-review-r2.md) - [proposal-review

- [Vision Skill](../docs/proposals/2026-04-29-vision-skill.md) - [Vision Skill Quality Refinement](../docs/proposals/2026-04-30-vision-skill-quality-refinement.md) - [Vision Skill Simplification and VISION.md Migration](../docs/proposals/2026-05-01-vision-skill-simplification-and-vision-md-migration.md) - [Optimize Vision Skill Strategic Positioning Quality](../docs/proposals/2026-05-06-optimize-vision-skill-strategic-positioning-quality.md)

This skill depends on Codex agent config.

This skill depends on dollar-prefixed Codex skill invocation.

This skill documents only the Codex install location.

This skill documents Codex and generated adapter install locations.

This skill requires Codex invocation wording.

This skill assumes Codex runtime controls.

This skill uses generic RigorLoop artifact paths.

This skill has valid metadata but invalid Markdown instruction content.

This fixture fails portable description validation.

This skill has an invalid portable name.

This skill is compatible with Codex and Claude Code, but not opencode.

Use generic RigorLoop artifacts without tool-specific assumptions.

Use a packaged asset while remaining portable across adapter roots.

This skill has Codex frontmatter that can be removed for other adapters.

This skill has unsupported metadata for non-Codex adapters.

generated fixture

First duplicate-name fixture entry.

Second duplicate-name fixture entry.

This fixture is invalid because its frontmatter omits `description`.

Invalid fixture skill missing the required Expected output section.

Invalid fixture skill missing the required name field.

Valid sibling skill used to prove the validator still catches a neighboring missing SKILL.md.

Invalid fixture skill missing the required top-level title.

Invalid fixture skill that still contains placeholder text.

Build, refresh, or audit a current-state repository orientation map with cited evidence, bounded inference, known gaps, and durable root or area scope.