codex-handoff
Skill0 starsChanged 5 months agoMITView on GitHub
Hand off a project or task to the OpenAI Codex CLI with full orchestration — loads the user's Codex config, applies OpenAI's official best practices, builds a tiered multi-agent execution plan, and runs the delegate-and-verify loop. Invoke when the user types /codex-handoff, or asks to hand a task/project over to Codex, run a Codex-orchestrated build, or delegate substantial work to their OpenAI tokens.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- ParaGenie/claude-codex-handofftheir own version, first seen Sep 6, 2026
- zjp1997720/zhijian-skillstheir own version, first seen Jul 17, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- zjp1997720/zhijian-skills★ 787same name
- ParaGenie/claude-codex-handoff★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repositoryOne other file it gives its agents.
- codexSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 1 is distinctive, and 7 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
7
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 15
Companion skill: read the companion codex skill (installed alongside this one, e.g.
- Very specific: line 16
User's live Codex config: read ~/.codex/config.toml (at minimum: model, model_reasoning_effort, sandbox_mode, [features], MCP servers).
- Very specific: line 17
If sandbox_mode is permissive (e.g.
- Very specific: line 20
Check available models in ~/.codex/models_cache.json before hardcoding tier choices (gpt-5.5 also has a "Fast" priority service tier — 1.5x …
- Very specific: line 22
OpenAI best practices: read references/openai-best-practices.md in this skill for the distilled official guidance.
- Very specific: line 33
Follow the companion skill's delegate-and-verify loop exactly: background runs, -o result files, structured output for fan-outs, …
- Distinctive: line 37
Run the full test/build yourself over the final state.
- Very specific: line 38
If durable Codex-relevant conventions emerged (commands, layout, prohibitions), offer to persist them to the project's AGENTS.md so every …
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
/codex-handoff — Orchestrated handoff to Codex
You are the tech-lead orchestrator. Codex workers (codex exec background runs) do the heavy lifting on the user's OpenAI tokens; Claude subagents fill gaps (Sonnet for workers, Opus for verification); you plan, dispatch, verify, and synthesize — you do not do worker-level work in the main loop.
The argument is the task. If no argument was given, ask what to hand off.
Step 1 — Load the environment (do this first, in parallel)
- Companion skill: read the companion
codexskill (installed alongside this one, e.g.~/.claude/skills/codex/SKILL.md) and itsreferences/orchestration.md— they define invocation flags, model tiers, briefing structure, and the phase playbook. This skill builds on them; don't duplicate, follow them. - User's live Codex config: read
~/.codex/config.toml(at minimum:model,model_reasoning_effort,sandbox_mode,[features], MCP servers). Key implications:- If
sandbox_modeis permissive (e.g.danger-full-access), always pass an explicit-s workspace-writeor-s read-onlyper run. - If
features.multi_agent = true, Codex has native subagents — for a big self-contained chunk you can give ONE Codex run a project-manager-style brief and tell it to use its own subagents for exploration/testing, instead of micro-managing many small runs. Prefer this when subtasks need shared context; prefer your own fan-out when subtasks are independent. - Note configured MCP servers/plugins (e.g. Cloudflare, browser, GitHub) — Codex workers can use them, so briefs may say "use your Cloudflare MCP to check the deployment".
- Check available models in
~/.codex/models_cache.jsonbefore hardcoding tier choices (gpt-5.5 also has a "Fast" priority service tier — 1.5x speed at increased usage — worth flagging when the user wants speed over cost).
- If
- Project ground truth: git status (warn if dirty before write runs), the project's AGENTS.md / CLAUDE.md, and the test/build commands.
- OpenAI best practices: read
references/openai-best-practices.mdin this skill for the distilled official guidance.
Step 2 — Plan the handoff
Produce a short written plan before spending tokens:
- Decompose into phases (Understand → Plan → Implement → Verify → Synthesize per the orchestration playbook).
- Assign each work item a worker (Codex tier or Claude subagent) and file ownership.
- State the fleet shape and rough cost posture to the user (e.g. "3 fast Codex scouts, 1 native multi-agent build run on gpt-5.5, Opus verification") — then proceed unless the task is destructive or scope is ambiguous.
Step 3 — Execute
Follow the companion skill's delegate-and-verify loop exactly: background runs, -o result files, structured output for fan-outs, cross-model verification (Opus reviews Codex's work; deep-tier Codex reviews Claude-written code), resume-with-error on failures, revert after two failed resumes.
Step 4 — Close out
- Run the full test/build yourself over the final state.
- If durable Codex-relevant conventions emerged (commands, layout, prohibitions), offer to persist them to the project's
AGENTS.mdso every future handoff starts smarter (codex /initscaffolds one). - Report: what each worker did, what you verified yourself, files changed, anything flagged, and approximate run count per tier.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

