agentleFS
schen22schen22/claude-config/skills/delegate/SKILL.md

delegate

Skill0 starsChanged 2 days agoMITView on GitHub

Use before EVERY Agent-tool call that starts an implementation or review helper (not Explore, claude-code-guide, Plan or a fork). Decides inline vs helper, sizes and splits lanes, and gives the brief template (Goal, Owns, Snippets, Gate, Report) so a helper doesn't start from zero or burn tokens on logs and screenshots.

86 other projects use a version of this.

Adopted by

Most recently taken up by

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.
  • MicrosoftSkill

    state

    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

  • kv0906Skill

    preview

    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

  • GitHubSkill

    prd

    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

  • nicolasestremSkill

    document

    Create focused documentation for specific components or features

    Also in 3 projects that use this

  • anombyte93Skill

    discover

    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 AzureSkill

    skill-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

The file

Files
SKILL.md

Delegate

Cost ≈ context size × steps. Everything a helper reads at startup rides along on every later step, so a 60-step lane that read 20k of docs pays for it 60 times.

Decide

One fresh helper per task keeps every context small; one helper doing many tasks carries all of them on every step (lane H: 5 items, 202k peak, 15.4M tokens).

SituationDo
Small item (≤ ~2 commits), or the orchestrator already has the files loadedInline — a helper's startup (~25–35k) costs more
A task that owns files no one else touchesOne fresh helper per task (1–2 items). Never more than 2 items per helper
Several such tasksRun them one after another; each brief pastes the previous report. Parallel only when they share no files
Sarah raises something independent of the main thread (feedback, a side fix, a review)A background helper (own worktree if it edits), so the main thread keeps going; relay its report when it lands. It never stops, replaces or re-scopes running threads or helpers unless Sarah says so
Follow-up on a finished helper's work (a blocker, a small fix)SendMessage the same helper (keeps its context)
Reviewreview-gates: Gate 1 inline per task; a reviewer helper only for risky tasks and once per batch

Brief template

Paste, don't point. The helper reads nothing beyond this brief and the files it owns.

## Goal
<one sentence; PLAN.md item ids>

## Owns
<files / regions it may change; nothing else>

## Snippets
<pasted excerpts it needs — the current code to change, the rules that apply (tokens,
CLAUDE.md, design docs, source tiers), a sample of the expected output. Aim ≤ ~3k tokens.
Any heading works (`## Rules`, `## Rows`). No "read CLAUDE.md" / "read docs/…".>

## Gate
- While iterating: <fast gate, e.g. npm run gate:fast>
- Before each commit: <full gate, e.g. npm run gate>
- Never run raw test commands or paste logs; the gate scripts filter output.
- Prove "fails before" with the single test (name filter), never the full suite.
- Never `git stash`: every worktree shares one stash list. For "fails before":
  `git diff <file> > <scratch>.patch`, `git checkout <file>`, test, `git apply` it back.
- Cap any other output (`| head -c 4000`). Never print generated files (builds, lockfiles).
- No screenshots. Verify visuals by hash or computed values; the visual call is Sarah's.
- Commit on lane/<name> in your worktree. Never push, never merge.

## Report (≤ 15 lines)
Commits (hash + subject), git diff --stat, gate result, what's left, anything Sarah
must look at. No logs.

Read-only helpers (reviews, research, ratings — nothing to commit): start with ## Read-only, then only Goal, Decisions, Snippets and Report. No Owns or Gate. ## Decisions is required: paste the PLAN Decisions the review touches (or "none apply"), so a finding that would reverse one is flagged as such, not proposed as a fix.

Pass model: "sonnet" (or "haiku" for purely mechanical edits). For opus or fable, add ## Why this model: "Sarah asked", or the judgement only that model can make.

While it runs

  • No git stash in your own tree either while a helper's worktree exists — a pop can take the helper's entry.
  • A helper past ~150k tokens: SendMessage it to commit what's green, report, and stop. Re-brief the rest as a new, smaller lane — a fresh brief is cheaper than every further step at 150k+.

After

  • The completion notice shows the helper's tokens. Over ~60k → add a Lessons line with the cause (see ~/.claude/CLAUDE.md § Where a lesson goes).

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.