codex-delegation
Skill1 starsChanged 2 months agoMITView on GitHub
Delegate a scoped task to Codex, prepare a Codex handoff, or manage an existing Codex plugin job. Use for explicit Codex delegation, rescue, delegated implementation or investigation, and Codex job status, results, or cancellation. Do not use for ordinary local work, generic model advice, Codex product questions, or structured autoreview.
3 other projects use a version of this.
Adopted by
Most recently taken up by
- aikazu/claude-codex-mcptheir own version, first seen Nov 8, 2026
- salalaslam/skillstheir own version, first seen Oct 20, 2026
- swjturay/codex-agy-delegatortheir own version, first seen Jul 25, 2026
Other versions3 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 3
- swjturay/codex-agy-delegator★ 4same name
- aikazu/claude-codex-mcp★ 0same name
- salalaslam/skills★ 0same 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 repository9 other files it gives its agents.
- AGENTS.mdAGENTS.md
- autoreview/AGENTS.mdAGENTS.md
- autoreviewSkill
- goagoSkill
- model-selectionSkill
- plansSkill
- starportSkill
- technical-writingSkill
- use-modern-goSkill
The file
We split this skill into the rules it gives an agent
12 rules. 0 are common, 12 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
12
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 9
Keep task ownership in the parent session.
- Distinctive: line 15
Read the repository instructions and inspect the worktree status.
- Distinctive: line 19
Run its assignment check before dispatch:
- Distinctive: line 25
Check the selected host's current model catalog and supported effort values.
- Distinctive: line 33
Use the checker for worker, reviewer, coordinator, and specialist assignments.
- Distinctive: line 47
Do not invent plugin commands, model switching, or message channels for that host.
- Distinctive: line 49
If the native plugin cannot express the task, read the companion constraints before direct companion use.
- Distinctive: line 51
If an existing local stop wrapper applies, read the conditional stop-hook note.
- Distinctive: line 67
Use the native completion result for foreground work.
- Distinctive: line 71
Never infer job state from a timeout.
- Distinctive: line 77
Do not ask a delegated worker to commit early.
- Distinctive: line 80
Keep pushes, pull requests, deployments, and merges within the user's authority.
“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Codex delegation
Keep task ownership in the parent session. Delegate only work that the user or active workflow authorizes. Preserve the distinction between investigation and implementation.
Prepare the assignment
- Read the repository instructions and inspect the worktree status.
- Define the task, allowed files, public behavior, acceptance evidence, and stopping condition.
- Select the model and effort with the installed
model-selectionskill. Run its assignment check before dispatch:
"${AGENTS_HOME:-$HOME/.agents}/skills/model-selection/scripts/model-selection" check --model gpt-6.1-sol --effort high --role worker- Check the selected host's current model catalog and supported effort values.
- Pass the accepted model and effort through the host's native parameters.
The assignment check validates policy. It does not prove host availability or change the active session model. If the host cannot express the accepted pair, select another eligible pair or report the limitation. Do not silently change the model or effort.
Use the checker for worker, reviewer, coordinator, and specialist assignments.
It rejects Haiku and Fable workers or reviewers. Astra requires a focused,
narrow, highly complex specialist task. Opus coordination or highly complex
work requires multi-agent Ultracode. Read the installed model-selection skill
for its current rules and exception flags.
Dispatch
For Claude Code, read the native plugin procedure
before the first dispatch. Use openai/codex-plugin-cc through
codex:codex-rescue or its slash commands. Do not write a custom shell wrapper
or launcher.
For another host, inspect its exposed tools and model parameters before dispatch. Do not invent plugin commands, model switching, or message channels for that host.
If the native plugin cannot express the task, read the companion constraints before direct companion use. If an existing local stop wrapper applies, read the conditional stop-hook note.
For browser work, use the owner's existing permissions for every site the task requires. Read the browser procedure.
Give each worker one concept or independent task. Include repository paths, constraints, checks, and the required result. For long prompts, save the task in a file and tell Codex to read that file first.
Require the worker to report changed files, check commands, results, and unknowns.
Require a manual repository-grounded audit at the end. A Codex job never runs
the structured autoreview helper or nested reviewers.
Manage and verify
Use the native completion result for foreground work. Do not poll a foreground job. For an explicit background job, use the host's status and result interfaces when needed. Do not create a liaison subagent to watch it.
Never infer job state from a timeout. File modification times can show worktree activity. A command log can stay stale while work lands. Neither proves that a job finished. The plugin has no mid-turn inbox for a liaison to query.
The parent session independently verifies the diff and required checks before delivery. It owns commits, salvaged work, structured autoreview, and publication. Do not ask a delegated worker to commit early. Run required structured autoreview in the parent session before a push or pull request.
Keep pushes, pull requests, deployments, and merges within the user's authority. Report the verified result and remaining uncertainty to the user.
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.

