Skill Claude CodeCodex
Reports silent decisions in agent-authored work: an independent read-only pass lists every choice the spec or request did not prescribe, in plain language, least-confident first. Not a code review, not a PR body, and it cannot change code or block merge.
Skill Claude CodeCodex
Runs a lean, risk-scaled code review: one generalist pass first, specialists only when triggered by evidence, and external debate only for high-risk escalation.
Skill Claude CodeCodex
Runs the complete development workflow from session context to reviewed pull request: write a spec, run spec-review and revise the spec, create an implementation task plan with the harness-native task system, drain that queue, commit, publish a PR, run self-review-loop, and post /assisted-review-heavy. Use when the…
Skill Claude CodeCodex
Reads unresolved GitHub PR feedback, independently validates and groups it by owning seam, applies ranked remove/simplify or seam fixes for proven issues, pauses design-expanding or same-seam stacked changes, verifies the result, and replies with a reasoned disposition.
Skill Claude CodeCodex
Rewrite agent-drafted text so a human can skim it without AI jargon or bloated terminology. Use when editing PR descriptions, review comments, digests, specs, decision ledgers, interactive-review answers, or any other text people must read. Surgical: fix contaminated sections only; preserve technical claims…
Skill Claude CodeCodex
Opens a localhost interactive diagram of a pull request so a developer can click components and ask follow-up questions answered by the same agent session. Use when the user wants an interactive review, to explore a PR visually, to drill into a /pr-figure-style overview, or says /interactive-review. Not for posting…
Skill Claude CodeCodex
Interview users to turn an issue, rough plan, or implementation idea into spec-ready decisions. Use when the user asks to interview/probe/question a plan, when issue-to-spec needs design clarification, or before writing a spec that depends on product, technical, rollout, or scope decisions.
Skill Claude CodeCodex
Orchestrates the full investigation-to-spec workflow starting from a GitHub issue. Phase 1: explore the issue and codebase to build context. Phase 2: interview the user from a problem-solving perspective. Phase 3: author a spec and publish it to docs/specs/. Phase 4: assess complexity and conditionally launch the…
Skill Claude CodeCodex
Loads full context of a GitHub PR (diff, description, linked issues, commit history, review threads, CI status), produces a structured narrative summary grouped by logical concern, and stays ready for follow-up Q&A.
Skill Claude CodeCodex
Generates one LLM reviewer diagram for a GitHub pull request, hosts it as a GitHub user-attachment without committing, and posts it as a comment or returns markdown for a PR body. Use when the user asks for a PR figure, reviewer diagram, image on a PR, a picture of the change, or when pr-publish needs the figure.…
Skill Claude CodeCodex
Publishes finished work as a GitHub pull request: discovers branch state, drafts or refreshes a layered human-readable PR description (with required humanizer pass and a reviewer figure), pushes safely, and creates or updates the PR.
Skill Claude CodeCodex
Publish existing code-review findings as one GitHub PR review with inline comments (required humanizer pass in review-comment mode). Use when the user says /publish-review, publish the review, leave these as PR comments, post this review on the PR, drop these in as inline comments, or comment on the PR with these…
Skill Claude CodeCodex
Iteratively reviews a PR with fresh read-only reviewers, applies only ranked design-quality fixes for demonstrated blockers, pauses same-seam or mechanism-expanding work for decisions, and pushes one final squashed commit only after a clean review without unexplained complexity growth.
Skill Claude CodeCodex
Reviews a spec with risk-scaled specialist agents, deduplicates findings, and returns a binary APPROVED / REVISIONS NEEDED verdict.
Skill Claude CodeCodex
Write clear instructional and reference prose for documents a reader follows or looks facts up in: runbooks, procedures, how-to guides, READMEs, API or configuration references, troubleshooting pages, migration guides, release notes, architecture decision records, or long follow-along PR sections. Covers imperative…
Skill Claude CodeCodex
Authors a human-first design spec: a plain-language narrative reviewers can absorb in minutes, backed by an implementation appendix with the exhaustive detail agents need. Use when the user asks to write a spec or design doc from a discussion or investigation, or to rewrite an existing spec that is too verbose for…