skills/codex-exec-review/SKILL.md
Use before presenting implementation plans and after non-trivial commits that require review. Runs codex exec in read-only review mode, handles resume, MCP transport errors, and review timeout semantics.
npx skillsauth add lilpacy/dotfiles codex-exec-reviewInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
codex exec to review the plan.codex exec to review the committed change.command codex exec ..., not codex exec .... This
user's shell has alias codex='codex --yolo'
(~/dotfiles/common.sh), which silently overrides --sandbox —
verified empirically: a run through the alias showed
sandbox: danger-full-access even with --sandbox read-only passed.
command codex bypasses the alias. Check the run header's
sandbox: line to confirm read-only actually took effect.--sandbox read-only.codex exec resume, --sandbox is not available. Use
-c 'sandbox_mode="read-only"' and check the run header. If the
header is anything other than sandbox: read-only, stop using that
resumed run for the required review and rerun the review with a fresh
command codex exec --sandbox read-only ... invocation.-c model_reasoning_effort=medium only for review runs.-c service_tier=fast and -c features.fast_mode=true when appropriate.--skip-git-repo-check only when the review must run outside a Git repository.codex exec or external review command.command codex exec \
--sandbox read-only \
--model gpt-5.4 \
-c model_reasoning_effort=medium \
-c service_tier=fast \
-c features.fast_mode=true \
"このプランをレビューして。read-only sandboxなのでテスト・build・format・install・生成コマンドは実行せず、差分・設定・既存ログの読取だけで判断して。不足する実行結果があれば質問して。別の codex exec や外部レビューコマンドは絶対に起動しないで。瑣末な点へのクソリプはしないで。致命的な点だけ指摘して。回答内容が現時点で out of date / deprecated になっていないかにも気をつけて: {plan_full_path} (ref: {CLAUDE_md_full_path})"
Use the same session for updated plan reviews:
command codex exec resume \
-c 'sandbox_mode="read-only"' \
--model gpt-5.4 \
-c model_reasoning_effort=medium \
-c service_tier=fast \
-c features.fast_mode=true \
<SESSION_ID> \
"前回の指摘を反映してプランを更新した。もう一度レビューして。read-only sandboxなのでテスト・build・format・install・生成コマンドは実行せず、差分・設定・既存ログの読取だけで判断して。不足する実行結果があれば質問して。別の codex exec や外部レビューコマンドは絶対に起動しないで。瑣末な点へのクソリプはしないで。致命的な点だけ指摘して。新しく追加された問題がなければ、その旨を明示して: {plan_full_path} (ref: {CLAUDE_md_full_path})"
After starting a resumed review, read the run header before waiting on
the final answer. Treat sandbox: workspace-write, sandbox: danger-full-access, or a missing sandbox line as review setup failure,
even if the reviewer was instructed in the prompt to behave read-only.
codex exec may print MCP transport errors such as http://127.0.0.1:8000/mcp connection failures at startup.codex exec failure.codex response is returned after the error, consider the run successful but degraded.codex exec exits without a final codex answer or without the requested review result.codex exec failure.review started but final result not yet returned is not review complete.codex exec resume -c 'sandbox_mode="read-only"' <SESSION_ID> or rerun
with a fresh command codex exec --sandbox read-only ....review incomplete and ask the user how to proceed.data-ai
ユーザー向けの回答、要件整理、説明、計画、レビュー、仕様、要約を作成するときに使う。自然言語だけでは条件、状態、多重度、期間、時刻、境界値、制約、推論、計算が曖昧になりうる場合、表、デシジョンテーブル、Mermaid 図、計算式、境界値表で一意化する。推測で補った図表要素は必ず明示する。
tools
Cross-agent messaging via SQLite. Send messages between Claude Code, Codex, Gemini CLI, and other agents. No daemon, no network, no dependencies beyond bash and sqlite3.
development
Anti-AI-slop design skill for greenfield pages, audits, redesigns, and design extraction from URLs or screenshots. Use when the user asks to build a new app or landing page, wants to redesign something, invokes Hallmark by name, or uses audit/redesign/study.
development
Reviews animation and motion code against a high craft bar derived from Emil Kowalski's design engineering philosophy. Default to flagging; approval is earned.