Skip to content

NickolasBenakis/agent-playbook / skills/delegate

delegate

Unverified

Decide whether to hand work to a subagent or child thread, and write the brief - what context to pass, what to leave out, and what to ask back. Use before launching any subagent, reviewer, or bb child thread.

Delegate

A helper knows only what you put in the brief, and you get back only its summary. Both handoffs lose detail. Delegate when that loss buys something: time, a clean context, or an unbiased opinion.

Pick the vehicle

NeedUseWhy
Find something across many filesan Explore/read-only subagentyou keep its conclusion, not the file dumps
A second opinion on a diffplaybook-reviewer with the criteria and a diff fileit can't be anchored on your reasoning
Several independent reads or reviewsparallel subagents in one messagecheap; they finish together
Read-only work over 3+ separate areas, too much to read yourselfone subagent per area, launched togetheryou keep conclusions, not dumps
Independent builds that each run 10+ minutes of agent timea child thread or worktree subagent per buildonly then does parallel time beat start-up cost
A small build (most fixes take this model 1–2 min), or one that shares files with yoursdo it yourselfthe handoff costs more than the work

Two builds that touch the same files are one task: give them to one agent.

The brief

Write it as if for a capable colleague who just walked in:

  1. Goal and finish line, as checks ("npm test passes, GET /x?page=2 returns the second page").
  2. Pointers, not payloads: file paths with line ranges, the commands to run, the branch to start from. Don't paste file contents or the chat.
  3. What's settled: decisions already made, approaches ruled out and why, so it doesn't redo them.
  4. What to reuse: existing helpers, tables, patterns and test setup, by path. Children that weren't told reinvented existing code.
  5. Constraints: scope limits, and the playbook rules that bite here (Decisions, secrets, no push to the default branch).
  6. What to return: at most ~15 lines: the result, the evidence (command and its outcome), Decisions, open questions. Diffs as a branch or file path, not inline.

Never put secret values in a brief; name the variable.

For a verifier or reviewer, pass the task and the artifact only, not your hypothesis or your reasons: a helper handed a conclusion tends to confirm it.

When it comes back

A summary can be confidently wrong. Re-run the finish-line check yourself before you report the work as done, and read the diff of anything you will put your name on.

reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 467kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 319kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 319kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 319kUpdated Claude CodeCodexVerifiedInstall