delegate
Skill4 starsChanged 32 days agoMITView on GitHub
Orchestrate cheaper or specialised worker models so the orchestrator (Claude Code or Codex) spends its tokens on planning, decisions and verification instead of bulk reading or mechanical edits. Use when a task involves reading many files or long logs/docs, exploring an unfamiliar codebase, mechanical multi-file edits, boilerplate, tests from a spec, a second-opinion review, commit messages or summaries. Workers include Claude Code, Codex, Antigravity (Gemini), Cursor, GitHub Copilot, Gemini CLI, OpenCode, Qwen Code, custom CLIs and any OpenAI-compatible endpoint (Ollama, LM Studio, OpenRouter…). Also trigger on "delegate", "use a cheaper model", "second opinion", "save tokens", "başka modele ver", "ucuz modelle yap", "ikinci görüş", "token tasarrufu".
- Reads credentials
85 other projects use a version of this.
Adopted by
Most recently taken up by
- BoredHF/albertskillstheir own version, first seen Nov 16, 2026
- finolabs/finolabs-vibecodetheir own version, first seen Nov 16, 2026
- timursafiullin/aodtheir own version, first seen Nov 15, 2026
- B16B1RD/bb-writtheir own version, first seen Nov 15, 2026
- hyprcat/skillstheir own version, first seen Nov 15, 2026
- finolabs/vibecodetheir own version, first seen Nov 15, 2026
Used with
Projects with it also use state, preview and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
MicrosoftSkillstate
Store and query a Microsoft Managed App's built-in state and file attachments with `ms project state` CLI schema commands and generated TypeScript clients. First, with the user signed in to the CLI, run `ms feature status --name state --json`; if `feature.enabled` is not `true`, recommend an alternative connector or data source.
Also in 3 projects that use this
kv0906Skillpreview
Render vault notes as styled HTML and open in browser. Turn markdown docs, decisions, progress reports into readable previews. Use for "/preview docs/project-a/checkout-flow.md", "/preview progress all", "/preview decide auth-approach", or when user wants to see a doc before sharing.
Also in 3 projects that use this
GitHubSkillprd
Generate high-quality Product Requirements Documents (PRDs) for software systems and AI-powered features. Includes executive summaries, user stories, technical specifications, and risk analysis.
Also in 5 projects that use this
nicolasestremSkilldocument
Create focused documentation for specific components or features
Also in 3 projects that use this
anombyte93Skilldiscover
Phase 1 of the prd-taskmaster pipeline: brainstorm-driven discovery. Delegates to superpowers:brainstorming in Interactive Mode (one adaptive question at a time), or self-brainstorms in Autonomous Mode when no user is present. Intercepts before the brainstorming chain hands off to writing-plans — this skill owns the exit. Extracts constraints, calibrates scale (Solo / Team / Enterprise), and advances the pipeline to GENERATE.
Also in 3 projects that use this
Microsoft AzureSkillskill-authoring
Write Agent Skills that comply with the agentskills.io specification. WHEN: "create a skill", "new skill", "write a skill", "skill template", "skill structure", "review skill", "skill PR", "skill compliance", "SKILL.md format", "skill frontmatter", "skill best practices". DO NOT USE FOR: general documentation. INVOKES: vally CLI.
Also in 3 projects that use this
Other versions85 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 85
- prassanna-ravishankar/repowire★ 270same name
- wlsdks/ontology-atlas★ 156same name
- HurricaHjz/second-yourself★ 150same name
- HurricaHjz/SecondSelf★ 150same name
- spmtitsolution/OpenBA-Kit★ 38same name
- pvjbuilds/delegate-router★ 32same name
- PyModel/claude-architect★ 31same name
- peterpme/skills★ 28same name
- jisuzzzz/haechi★ 17same name
- mhlaghari/stuntman★ 16same name
75 more not listed
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.
The file
We split this skill into the rules it gives an agent
7 rules. 0 are common, 3 are distinctive, and 4 are very specific to this project.
0
Common
Most of the 74 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 27
Read-only is the default. --write lets the worker edit files under --dir (default: cwd), but it can never commit or push.
- Distinctive: line 44
Endpoints (text only, no file access): commit messages, summaries, translation.
- Distinctive: line 46
Do it yourself: when the task needs tight back-and-forth, or touches secrets, auth, payments or infrastructure.
- Very specific: line 80
Run: Run independent tasks as parallel delegate calls in the background when your host supports it.
- Very specific: line 85
Escalate instead of looping: If a worker fails twice on the same task, do it yourself or use a stronger worker or model (--model).
- Very specific: line 93
Run only one --write worker per directory at a time.
- Distinctive: line 116
If the user asks what workers did or what usage costs, point them to the dashboard at http://127.0.0.1:4747.
“Common” is measured: kept by at least 6 in 10 of the 74 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
Delegate: you orchestrate, workers execute
You are the orchestrator. You own the plan, every decision and the final verification. Workers
are other agents and models that you drive with the delegate command. Their full transcript
goes to a log file; only their ## REPORT, which leads with the answer or findings, reaches your
context.
The user installed conductor so that you hand work to these workers. Using delegate counts as
the user asking for it; it is not the Agent tool. When to use it is your call.
Which delegate: in Claude Code, delegate is on PATH. In Codex and other hosts, use the
full path printed at session start (…/bin/delegate).
delegate <worker> [--write] [--model M] [--timeout SEC] [--dir D] "<task>"
git diff --staged | delegate <worker> "write a one-line conventional commit message" # stdin = context
delegate list # usable workers and what each is good at (you are excluded)
delegate status # every known worker: enabled / installed / configured
delegate models <worker> # models a worker offersRead-only is the default. --write lets the worker edit files under --dir (default: cwd), but it
can never commit or push. After a --write run, delegate prints git status --short for you.
Routing
The session-start note lists the workers available here and what each is good at; the user sets
these on the dashboard's Workers page. If you lack that note, run delegate list. Pick by the
"good at" line, then by cost: free or self-hosted endpoints for text chores, the user's
subscriptions for agentic work. A worker marked "last run failed" may have an expired login, so
check its log before relying on it.
Typical fits:
- Reading and analysis: reading many files, mapping a codebase, summarising modules, long logs or docs.
- Implementation (
--write): well-specified, mechanical edits across files, tests from a spec, migrations. - Review: an independent review or second opinion from a different model family.
- Endpoints (text only, no file access): commit messages, summaries, translation. They are the least reliable, so never use them for decisions.
- Do it yourself when the task needs tight back-and-forth, or touches secrets, auth, payments or infrastructure.
When. The user sets how hard to delegate on the dashboard (Settings → Delegation level); the
session-start note carries that level as a concrete list. At the default level, delegate by
default when the next step is exploring or reading more than ~3 files, a long log or doc, a review,
a multi-file or mechanical edit, tests from a spec, translation or a commit message. Do it yourself
when it is a small edit in a file you already read, a decision, or anything with secrets, auth,
payments or infra. Do not weigh "is it worth it" case by case: a worker's report costs you a few
hundred tokens, reading the files yourself costs thousands. A mid-session conductor: reminder
after several inline reads means you have drifted; hand off the next chunk.
Models. A worker can have a model pool (dashboard → Workers): models switched on with a share
each. Without --model, delegate picks from the pool so the shares hold over time; delegate
list shows them. Pass --model when the task needs a specific one (a stronger model for a hard
task, a fast one for a lookup).
Protocol
Brief like a contractor: long brief, short report. The worker has none of your context, and a detailed brief is cheap for you while a vague one costs a retry. Send one task per call, in this shape (write long briefs to a file and pass
"$(cat brief.md)"):GOAL: what to produce and why, in one or two sentences FILES: what to read first; for --write, exactly which files it may change CONTEXT: what you already know (APIs, decisions made, conventions to match) so it need not rediscover it CONSTRAINTS: what not to touch, style, APIs to keep DONE WHEN: the acceptance checkWorkers end with a fixed, short report (under 120 words unless the user chose full reports):
STATUS: done|partial|failed,ANSWER:,CHANGED:,UNSURE:. Do not ask for more than that; the full transcript is in the log if you ever need it.- Run. Run independent tasks as parallel
delegatecalls in the background when your host supports it. - Verify. Never relay a worker's claim as fact. For
--write, read thegit diffof the touched files, run the project's tests and linters, and fix or revert what is wrong. For read-only answers, spot-check one or two of the cited locations before building on them. - Escalate instead of looping. If a worker fails twice on the same task, do it yourself or
use a stronger worker or model (
--model). - Tell the user which parts were delegated, to whom, and what you verified.
Rules
- Workers must not see secrets. Never pipe
.envfiles, keys, tokens orsecrets/intodelegate, and do not delegate tasks whose subject is credentials. - Run only one
--writeworker per directory at a time. Parallel--writework needs separate git worktrees (--dir <worktree>). - Workers cannot delegate further: nested
delegatecalls are refused with exit code 4. Exit codes:
Code Meaning 0 done 2 usage error or unknown worker 3 the worker is disabled, not installed or not configured 5 the worker said nothing (for example a tool was denied in headless mode) 6 the worker is out of quota or hit its plan limit: use another worker, do not retry it until it resets 7 the worker is not signed in (or its API key is invalid): ask the user to sign in 124 timeout 130 interrupted A worker that prints a quota, usage-limit or sign-in error is stopped within seconds instead of running to its timeout, and
delegate listmarks it, so the session-start worker list warns you off it until it resets.- The full transcript is at the
log=path on the first output line. Read it only when the REPORT is not enough. - For images, use
gen-image(theimageskill), not these workers. - If the user asks what workers did or what usage costs, point them to the dashboard at http://127.0.0.1:4747.
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.

