agentleFS
ymchoiymchoi/qbpowers/.claude/skills/worker/SKILL.md

worker

Skill3 starsChanged 4 days agoUnlicenseView on GitHub

Choose the workers this session delegates to. Usage /worker [claude-light | claude | hybrid-light [fast|ultrafast] | hybrid [fast|ultrafast] | codex-light [fast|ultrafast] | codex [fast|ultrafast] | codex-max [fast|ultrafast] | none], default hybrid.

11 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 2 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 10 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If a worker fails, report it.

    line 36
  2. Very specific:

    Run codex with the scripts in .claude/skills/worker/scripts/, using the repo .venv Python.

    line 44
  3. Very specific:

    Write each task to <WORK>/prompt.txt.

    line 47
  4. Very specific:

    Add --workroot <DIR> when codex must work outside the repo root, such as in a worktree.

    line 50
  5. Distinctive:

    Never give one work dir to two waiters, and never stop a waiter's background task: that also kills the codex runs it started.

    line 51

“Common” is measured: kept by at least 6 in 10 of the 10 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Worker routing

Until the session ends, delegate work by the column this invocation selects. The latest /worker wins, and a routing table the user gives in the session replaces this one until the next /worker.

  • /worker or /worker hybrid: the hybrid column.
  • /worker claude-light: the claude-light column.
  • /worker claude: the claude column.
  • /worker hybrid-light: the hybrid-light column.
  • /worker codex-light: the codex-light column.
  • /worker codex: the codex column.
  • /worker codex-max: the codex-max column.
  • /worker none: start no workers, Claude or GPT, and do all the work yourself. Workers already running finish; take their results.

fast or ultrafast after hybrid-light, hybrid, codex-light, codex or codex-max puts that column's GPT workers on service tier priority or ultrafast. Otherwise, and in both claude columns, GPT workers run at standard speed.

Workclaude-lightclaudehybrid-lighthybridcodex-lightcodexcodex-max
Hard: a bug of unknown cause, a design or architecture decision, a change across several modules, a change that is hard to undo (security, auth, payments, data migration)opus xhighopus maxopus xhighopus maxastra maxastra maxastra max
Writing or changing code, tests or documentsopus highopus xhighsol maxastra maxsol maxastra maxastra max
Deep review of an important changeopus highopus xhighopus highopus xhighsol maxastra maxastra max
Web search; research in external documents or open sourceopus mediumopus highsol maxsol maxsol maxsol maxastra max
General: the scope is set and judgment stays inside it. Reproduction, simple diagnosis, routine review and checksopus mediumopus highsol maxsol maxsol maxsol maxastra max
Simple: nothing to interpret. Finding, reading, listing; running tests and reporting the result; a spelled-out edit in one or two files; mechanical repetitionopus mediumopus mediumsol maxsol maxsol maxsol maxastra max
Computer Use, one such worker at a timesol maxsol maxsol maxsol maxsol maxsol maxastra max

opus is a Claude subagent; astra (gpt-6-astra) and sol (gpt-6.1-sol) run through codex. Computer Use is the exception in every column: it always runs through codex.

Whenever you can, split work across workers and run them in parallel; parallelize only tasks that can run side by side without trouble.

If a worker fails, report it; do not quietly hand its work to a different worker or do it yourself. Trying the same worker again is fine.

Claude workers

Spawn with the Agent tool: subagent_type: worker-effort-<effort>, model: opus.

GPT workers

Run codex with the scripts in .claude/skills/worker/scripts/, using the repo .venv Python.

  1. new_workdir.py [COUNT] prints COUNT new work dirs.
  2. Write each task to <WORK>/prompt.txt. For a JSON result, add <WORK>/schema.json.
  3. Hand the work dirs to one waiter in the background (run_in_background: true): wait_codex.py --model <astra|sol> --effort <effort> [--service-tier <tier>] <WORK>... Add --workroot <DIR> when codex must work outside the repo root, such as in a worktree. Add --cap N (runs at once, default 64), --launch-gap S (seconds between starts) or --retries N (fresh restarts of a run that failed for good) when the user asks for them. Never give one work dir to two waiters, and never stop a waiter's background task: that also kills the codex runs it started. To act on each run as it ends, give each its own waiter.
  4. Act on the summary:
    • OK <file>: codex's final message.
    • FAILED <WORK> <kind>: report it; do not rerun blindly.
    • STALLED <WORK> <minutes>: report it; never kill codex for being slow. The same command waits another round.
    • RUNNING / QUEUED: the same command continues the batch. After a usage limit, rerun it once the reset time shown has passed.
  5. To answer a worker's question or give it follow-up instructions, write them to <WORK>/reply.txt and run the waiter on that work dir again. The run continues in the same codex session.

The waiter makes schema.json strict (additionalProperties: false, every property required). Avoid what it cannot fix: a non-object root, allOf/not/if/then/else/dependentRequired/dependentSchemas, and maps (use arrays of key/value objects).

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.