Init codex config
Linlin's curated AI agent harness configuration: workflow rules, skills, hooks, plugin recommendations, tooling preferences, and project templates. Loadable into any new project so a fresh /init can pick the relevant subset.
npx -y skills add jajupmochi/agent-harness --skill init-codex-configAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Scaffold or migrate a project to use agent-harness from Codex. Use when applying agent-harness in Codex, converting Claude Code setup to AGENTS.md/.codex/.agents, or making a project ready for Codex skills and hooks without changing the original Claude setup.
SKILL.md
4.3 KB, as published. Nobody here has run it
init-codex-config
Apply agent-harness to a project for Codex. This is the Codex counterpart to
setup/init-agent-config/SKILL.md; read that source skill before adapting a
project, then translate only the selected pieces into Codex surfaces.
Preflight
- Locate this repository. Prefer the current plugin root; otherwise check
~/.claude/agent-harness. - Read:
../../docs/CODEX_ADAPTATION_PLAN.md../../INVENTORY.md../../rules/README.md../../setup/init-agent-config/SKILL.md
- Inspect the target project for existing
AGENTS.md,.codex/,.agents/skills,.claude/,CLAUDE.md, package manifests, and Git status. - Do not edit
.claude/orCLAUDE.mdunless the user explicitly asks for Claude Code changes too.
Decide the Codex surfaces
- Durable project guidance goes in
AGENTS.md, notCLAUDE.md. - Codex project hooks go in
.codex/hooks.json. - Repo-scoped Codex skills go in
.agents/skills/<skill-name>/SKILL.md. - User-wide Codex skills go in
~/.agents/skills. - Reusable distribution goes through
.codex-plugin/plugin.jsonand a marketplace entry. - MCP or app integrations belong in Codex MCP/app config, not in AGENTS.md.
Questions
If the user has not already provided answers, gather these decisions. In plan mode, use the available user-input tool. Outside plan mode, ask only the questions that affect file writes.
- Project type: Python research, static site, frontend app, or custom.
- Language policy: English, Chinese, bilingual, or leave unchanged.
- Context tags:
always,research-pkg,ui-project,static-site,ml-research,web-perf,image-or-video-work,docs-site,electron-or-desktop. - Which personal rules to include: output brevity, tool proactivity, no reread, writing style, human-readable output, autorun mode.
- Skill scope: repo
.agents/skillsor user~/.agents/skills. - Hook scope: project
.codex/hooks.json, user~/.codex/hooks.json, or no hooks. The tested local activation uses user hooks; do not also declare plugin hooks or the same command can run twice.
When the user says autorun, choose conservative defaults and continue unless
a file overwrite, dependency install, network call, or destructive operation
needs approval.
Compose for Codex
- Preserve any existing
AGENTS.md; merge additively or createAGENTS.md.bak.<date>before replacing. - Generate
AGENTS.mdwith concise instructions. Do not paste the full rule library. Include references to selected source rule files, for example:Rules source: ~/.claude/agent-harness/rules/phased-planning/RULE.md. - If project hooks are selected, create
.codex/hooks.jsonfrom this repository's roothooks.json, then rewrite every command to a stable absolute or git-root-based path. The root file is a source template, not an automatically discovered plugin hook. - If repo skills are selected, copy the relevant top-level Codex wrapper
skill directories from this repository into
.agents/skills/. - If a template is selected, use the corresponding template directory as a
source, but translate Claude-specific files:
CLAUDE.template.md-> summarizedAGENTS.md.claude/settings.template.json->.codex/hooks.json.claude/skills->.agents/skills
- Add a short README note only when the project already has a README and the user wants provenance recorded.
Verification
Run checks proportional to the files written:
jq emptyon every JSON file created or changed.python3 -c 'import tomllib; ...'for TOML files if present.rg -n "CLAUDE|\\.claude" AGENTS.md .codex .agentsto ensure Codex files do not accidentally claim to be Claude configuration, except in provenance text.git status --shortto report changed files.
Do not claim that hooks or skills are active in the current Codex task until they have been installed in a discovered location and a real lifecycle event or explicit invocation proves that task picked them up. Trusting a hook definition does not prove an already-running task rebuilt its lifecycle registry.