- name
SKILL.md
Die Skill-Datei bleibt im Repository des Herausgebers.
Fügen Sie den kopierten Befehl in ein Terminal ein. Er installiert diese SKILL.md in dem jeweiligen Agenten.
npx skills add yan-labs/yan-skills --skill codex -a cursor -yFügen Sie den kopierten Befehl in ein Terminal ein. Er installiert diese SKILL.md in dem jeweiligen Agenten.
npx skills add yan-labs/yan-skills --skill codex -a claude-code -yFügen Sie den kopierten Befehl in ein Terminal ein. Er installiert diese SKILL.md in dem jeweiligen Agenten.
npx skills add yan-labs/yan-skills --skill codex -a codex -yFügen Sie den kopierten Befehl in ein Terminal ein. Er installiert diese SKILL.md in dem jeweiligen Agenten.
npx skills add yan-labs/yan-skills --skill codex -a grok -ySenden Sie die kopierte Nachricht in einem Grok Bot-Chat. Der Bot ruft diese SKILL.md von GitHub ab und fügt sie den Skills hinzu, die alle Ihre Bots gemeinsam nutzen.
Install this skill: https://github.com/yan-labs/yan-skills/tree/HEAD/codexVom Herausgeber
用 Codex CLI 承接用户指定的代码分析、编辑、审查或并行代理任务。图片生成的当前流程见 imagegen/SKILL.md;旧版操作与实验记录保存在 references/image-experiments.md。
路由前提:按全局 CLAUDE.md §2,编码、修 bug、调研与 review 默认用 Terra(fleet-go new --to gpt --simple,见 agent-fleet),不手写 codex exec。本 Skill 只在用户明确要求直接操作 Codex CLI(自选 sandbox、codex review/apply/resume 等原生子命令、并行 worker 的 worktree 布置)时使用;简单派单不要先读本文件。
Never block the main conversation on a codex exec call. Always launch via Bash with run_in_background: true. The only exception is a trivial codex --version health check.
AskUserQuestion unless they explicitly ask to be prompted. Model is gpt-6.1-sol (the default in ~/.codex/config.toml; pass -m only when the user names another model in the current request). Defaults:
medium; low for single-file edits with clear boundaries. Never escalate to high/xhigh on your own (global §2); only when the user asks for it explicitlyread-only unless the task clearly needs edits (workspace-write) or network (danger-full-access)cat /tmp/codex-prompt-<tag>.md | codex exec --skip-git-repo-check \
--config model_reasoning_effort="medium" \
--sandbox read-only \
-C <workdir> 2>/dev/null
run_in_background: true. Record the returned shell id and a short tag (e.g. codex-review, codex-refactor-auth) so you can reference it later.codex-review (medium effort, read-only) in background." Then continue with other work or wait for user input. Do NOT sit and poll.2>/dev/null 压低 stderr 噪音;排查启动失败或调试 CLI 时保留错误输出。--skip-git-repo-check. Put all flags between exec and resume (if resuming).BashOutput (or Read on the captured log file) — do not re-run the command.codex resume <tag> → you will run echo "<new prompt>" | codex exec --skip-git-repo-check resume --last 2>/dev/null (no other flags on resume; session inherits model/effort/sandbox).Codex sub-agents compose cleanly. To run an agent team:
Bash call in the same message (parallel tool calls). Give each a distinct tag and, if they write, a distinct -C workdir or separate git worktree to avoid edit collisions.workspace-write Codex workers against the same files concurrently. Either serialize them, scope them to disjoint directories, or run each in its own git worktree.read-only workers, each with a different lens (security, perf, API design). Cheap and fully parallel.workspace-write worker implements, then a read-only worker reviews the diff. Sequential, not parallel.fleet-go new --to gpt --simple) check the diff against the brief (global §4.3). Do not pair a Codex worker with a Claude sub-agent as the reviewer.Default model: gpt-6.1-sol, which ~/.codex/config.toml already sets. Only add an explicit -m flag when the user asks for a different model by name in the current request.
Reasoning effort: medium (standard default) · low (single-file, clear boundaries). high and above are never chosen automatically; use them only on the user's explicit request.
Cached input is 90% off for 24h — reuse the same prompt prefix across workers when possible.
Do not ration Codex calls and set no turn, context or time cap (global §3). The owner's plan does not need protecting from extra workers; optimize for the right answer and run to the acceptance condition. A limit the user states explicitly always wins.
codex --version 或启动失败,先运行 codex doctor 并读取错误,最多重试一次;仍失败(含 402、登录失效、模型被拒、产物为空或含裸 tool-call 控制 token)就停下并如实告知用户,由用户决定。不许静默改派 Claude subagent,Claude subagent 也不得再派 Claude subagent。--full-auto and --sandbox danger-full-access: it is
their own single-user machine and they prefer agents to act rather than ask. Pick the
sandbox the task needs and run. Still disclose it — the one-line launch report
names the sandbox, so "no gate" never becomes "no visibility". Never use
AskUserQuestion for a sandbox flag.The skill used to describe exec as if it were the whole CLI. It is not. Commands that
change what you would reach for:
Two exec flags the recipes above should use more:
-o <FILE> / --output-last-message <FILE> — writes the agent's final message to a
file. Prefer this over scraping stdout: stdout carries progress chatter, and parsing
it is exactly the kind of silently-wrong extraction this workspace has been bitten by.--output-schema <FILE> — a JSON Schema constraining the final response shape. Use it
whenever you need a structured result back, instead of asking for JSON in prose and hoping.Check with codex --version. Default model gpt-6.1-sol is configured in ~/.codex/config.toml — do not override it unless the user explicitly requests a different model.
本 Skill 当前位于 yan-skills/codex/;安装位置和符号链接以实际工作树为准。
codex exec in the foreground and making the user wait.AskUserQuestion before every launch — decide from context.workspace-write workers on overlapping paths.2>/dev/null and flooding the main thread with thinking tokens.>-
---name: codexdescription: >----# codex/SKILL.mdFAQ
Laden Sie codex/SKILL.md vom Herausgeber herunter oder kopieren Sie den Pfad und öffnen Sie das Repository. PlugMCP listet den Skill und hostet die Datei nicht.
Der Herausgeber pflegt es unter https://github.com/yan-labs/yan-skills.
Claude, Cursor, Codex und Grok Build können eine SKILL.md lesen, die Sie aus dem Repository des Herausgebers hinzufügen. Grok Bot installiert sie, wenn Sie den Link zum GitHub-Ordner des Skills in einem Chat senden.