codex-routing
Skill3 starsChanged 5 months agoMITView on GitHub
Use Sol as the advisor and coordinator for coding work, delegating investigation and implementation to Luna subagents.
In this repository8 other files it gives its agents.
- AGENTS.mdAGENTS.md
- appstore-readiness-auditSkill
- continue-deep-researchSkill
- deep-code-reviewSkill
- git-clean-merged-branchSkill
- research-repo-technologySkill
- swift-code-reviewSkill
- triage-review-commentsSkill
The file
We split this skill into the rules it gives an agent
10 rules. About 0 are common, 10 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈10
Distinctive · estimate
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 16
Use Luna High for clear, straightforward, strongly verifiable work.
- Distinctive: line 18
Use Luna xhigh for harder work that needs more investigation or reasoning, and as the default when the appropriate effort is uncertain.
- Distinctive: line 20
Use Luna Max only for exceptionally difficult, broad, or exhaustive bounded work.
- Distinctive: line 24
When a task has multiple meaningful independent pieces, fan the work out across several Luna workers.
- Distinctive: line 26
Look for natural workstreams such as separate code areas, implementation components, tests, regression checks, reviews, or unfamiliar …
- Distinctive: line 28
Prefer a fresh worker for a distinct responsibility.
- Distinctive: line 30
Use available subagent capacity when it is useful.
- Distinctive: line 50
If work is wrong or incomplete, send a focused correction to a Luna worker.
- Distinctive: line 54
Use Sol for judgement, decomposition, coordination, and review.
- Distinctive: line 56
Use Luna for execution.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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 routing
Sol thinks and coordinates. Luna executes. Sol reviews.
Sol should understand the request, decide the approach, split the work into sensible pieces, delegate those pieces to Luna subagents, review their results, coordinate corrections, and produce the final answer.
Sol should continuously coordinate the task, creating new Luna workers as new independent work becomes available. It should avoid taking over implementation work that can be delegated.
Delegation and parallelism
Use Luna High for clear, straightforward, strongly verifiable work.
Use Luna xhigh for harder work that needs more investigation or reasoning, and as the default when the appropriate effort is uncertain.
Use Luna Max only for exceptionally difficult, broad, or exhaustive bounded work. Expect Max workers to take longer.
Choose effort from each worker's bounded assignment, not from the size or difficulty of the overall task. A large multi-stage plan does not make every lane a Luna Max lane. Split broad work into useful assignments first, then use the lowest effort that is appropriate for each assignment.
When a task has multiple meaningful independent pieces, fan the work out across several Luna workers. Parallelize independent investigation and execution; keep only genuinely dependent work sequential.
Look for natural workstreams such as separate code areas, implementation components, tests, regression checks, reviews, or unfamiliar dependencies.
Prefer a fresh worker for a distinct responsibility. Do not keep sending unrelated new work to the same worker merely because it already exists.
Use available subagent capacity when it is useful. Do not force a small task into multiple workers when delegation would add no value.
Waiting for workers
Once a worker is running, wait for it to finish or report that it needs attention. Use long wait intervals appropriate to Luna, especially Luna Max, and continue waiting after ordinary wait timeouts.
Elapsed time, silence, or the absence of a patch is not evidence that a worker has stalled. Do not interrupt, replace, or duplicate a running lane for those reasons.
If a worker needs attention, handle or surface that need and then continue waiting. Replace a worker only when it explicitly fails, reports an unrecoverable blocker, terminates without completing the assignment, or the user directs a replacement. While waiting, Sol may coordinate other independent work, but it must preserve the original lane and eventually collect its result.
Assignments
Give each worker a clear, bounded goal and enough context to complete it successfully, including relevant files or components, important constraints, useful implementation context, and how to validate the result.
Let the worker reason about implementation details unless it must follow a specific design decision.
Review
Sol reviews worker output before considering the task complete. Check that the requested outcome was achieved, important constraints were respected, validation is credible, and no obvious regressions were introduced.
If work is wrong or incomplete, send a focused correction to a Luna worker. Sol remains the coordinator instead of taking over the implementation.
Principle
Use Sol for judgement, decomposition, coordination, and review.
Use Luna for execution.
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.

