菜单
Check AGENTS.md files for discovery, scope, precedence, contradictions, stale facts, executable commands, and native Codex boundaries. Use before merging instruction changes or diagnosing agent behavior.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Start here for the nddev-builder toolkit. Indexes the Codex artifact families and the creator, checker, reviewer, and workflow skill that owns each, points at the single reference for schemas and current (Beta, experimental, or under-development) surfaces, and routes the task to the right skill. Use first whenever building, checking, reviewing, or releasing a Codex artifact, or when unsure which builder skill applies.
A guided first run of the nddev-builder toolkit — from zero to a working, checked Codex artifact in ordered steps. Use when new to nddev-builder, onboarding, or unsure which builder skill to start with.
Create or revise a Codex marketplace catalog with valid plugin sources, categories, installation policy, and authentication policy. Use when publishing local or remote plugin collections.
Check a managed Codex requirements.toml layer for supported keys, permission-profile consistency, and a valid lower-layer default. Use before delivering managed /etc/codex or MDM requirements.
Create a managed Codex requirements.toml layer that constrains permission profiles and other admin policy. Use when authoring the /etc/codex or MDM-delivered managed requirements a lower config layer must satisfy.
Install and iterate on a Codex plugin locally in developer mode: the install, reinstall, and cachebuster test loop for a plugin under development, run against an isolated install and never live state. Use when running or iterating on a plugin, its skills, MCP, hooks, or bundled app locally.
| name | codex-instructions-checker |
| description | Check AGENTS.md files for discovery, scope, precedence, contradictions, stale facts, executable commands, and native Codex boundaries. Use before merging instruction changes or diagnosing agent behavior. |
Resolve every ../../scripts and ../../references path relative to this SKILL.md directory, never the caller working directory.
Audit the effective instruction chain and distinguish structural validity from semantic correctness.
Locate the target file and every applicable parent or child instruction file. Read ../../references/codex-artifact-contracts.md.
Run:
python3 ../../scripts/check_codex_artifact.py instructions AGENTS.md
Verify placement, filename, size, headings, referenced paths, and runnable commands.
Compare claims with code, config, manifests, and current repository state. Flag duplicated machine-owned values.
Review precedence for contradictions, accidental scope expansion, unsafe authority, and rules that belong in a skill or validator.
Reject copied slash-command models, collapsed foreign-CLI instruction files, secrets, speculation, and runtime-local state.
If diagnosing behavior, start an isolated Codex session from representative directories and inspect the discovered chain.
Return PASS only when structure, facts, and precedence are sound. Otherwise return FAIL with file and heading, conflicting evidence, scope impact, and correction.