MITupdated 21d ago
Use this skill when the user wants a staff-level review of a PR, diff, or code change.
Staff Code Review で何ができる?
name: staff-code-review
description: Review code changes with staff-level depth by reusing the source workflow under claude/staff-code-review. Use when the user wants a thorough review beyond line-level correctness.
metadata:
short-description: Review changes at staff level
Staff Code Review
Use this skill when the user wants a staff-level review of a PR, diff, or code change.
This is a thin Codex wrapper around claude/staff-code-review/skills/staff-code-review/SKILL.md. Reuse the source workflow, references, scripts, and eval material from the Claude skill directory instead of maintaining duplicated Codex copies.
Use This Skill
- Use this skill when the user wants a staff-level review of a PR, diff, or code change.
Do Not Use This Skill
- Do not use this skill for a quick syntax check or lightweight lint feedback.
Source Material
- Source skill:
claude/staff-code-review/skills/staff-code-review/SKILL.md - Source directory:
claude/staff-code-review/skills/staff-code-review/ - Codex metadata:
agents/openai.yaml
Load only the source files needed for the current task. Do not recreate or copy the Claude-side bundled resources into this Codex skill.
Workflow
- Treat
claude/staff-code-review/skills/staff-code-review/SKILL.mdas the source workflow. - Read the needed source references from
claude/staff-code-review/skills/staff-code-review/references/before escalating findings in those dimensions. - Ground the review in actual codebase context and report findings first, ordered by severity.
- The source workflow fans out to many subagents (Pass 1.5 Explore brief, seven dimension personas, the Code Adversary, Pass 2.5 translation, the Findings Adversary, and a humanize pass). Run them per the "Codex execution notes" below - inline by default - rather than copying Claude's
Agent/subagent_typecalls.
Codex execution notes - subagent spawning
The source workflow assumes Claude's Agent tool and named council:* subagent types, neither of which exists on Codex. It also fans out well past Codex's limits: a Research-Brief Explore agent, seven dimension personas, a Code Adversary, a per-persona translation pass, a Findings Adversary, and a humanize pass. Codex subagent fan-out is fragile at that size - as of mid-2026 the default agents.max_threads is 6, completed subagents do not free their slot unless explicitly closed (openai/codex#22779), and subagents can finish without returning their payload (#16051). Therefore:
- Default: sequential / inline. Do NOT spawn one subagent per dimension. Build the Research Brief yourself (grep/read), then work each dimension's checklist from
references/in turn in its own lens, then run the Code Adversary and the Findings Adversary as inline red-team passes. Thecouncil:*persona types are unavailable, so apply each lens directly and emit conventional-comment format - this makes Pass 2.5 (translation) a no-op, so skip it. Humanize inline. - Optional parallel mode (only if the user asks and has raised
agents.max_threads). Cap concurrency at <= 5, run the dimensions in waves,close_agenteach reviewer as soon as it returns (completion alone does not free the slot), and verify each returned a payload before synthesizing - re-run any missing one inline. Keep both adversary passes as final inline steps. Never assume all spawned reviewers reported. - Use native agent tools only for agent state (
spawn_agent/wait_agent/close_agent); do not run nestedcodex exec, and do not use shell (ps,pgrep,ls,rg) for agent orchestration.
Codex Notes
- Ignore Claude-only frontmatter and runtime wiring such as
allowed-tools,user-invocable,$ARGUMENTS, andCLAUDE_SKILL_DIR. - Infer inputs from the user request and local context before asking follow-up questions.
- If a source script or networked command fails because of sandbox restrictions, rerun it with escalation and a short justification.
- For destructive or irreversible actions, confirm intent unless the user was already explicit.
Verification
After any fix or state-changing action, rerun the narrowest relevant validator, script, listing command, or source check and compare the before/after result.
- Re-check the key output, diff, command result, or rendered text after you act.
- If the task changed files or external state, rerun the narrowest relevant validator, script, listing command, or source query.
- Report what was verified, what remains unverified, and any residual risk.
Examples
インストール
Staff Code Review をクライアントに追加します。お使いのものを選んでください。
npx skills add smykla-skalski/saiInstalls every skill in the repository, then prompts for which to keep.
/plugin marketplace add smykla-skalski/saiAdds the repository as a plugin marketplace; install individual plugins with `/plugin install`.
git clone https://github.com/smykla-skalski/sai
cp -r codex/staff-code-review ~/.claude/skills/A skill is a plain directory. Copy it into `.claude/skills/` in a project or in your home directory.
スコア
70 / 100
良好