AGENT SKILL

dispatching-parallel-agents

Use when facing 2+ independent test failures, bugs, research questions, map lanes, or implementation tasks that can run without shared write state. Dispatch native subagents through the owning sp-* workflow.

License 许可
MIT
最近更新
2026-08-07
Stars
★ 0
Forks
0

01 前言 Front Matter

name
dispatching-parallel-agents
metadata
{ "author": "github-spec-kit", "source": "templates/passive-skills/dispatching-parallel-agents/SKILL.md" }
description
Use when facing 2+ independent test failures, bugs, research questions, map lanes, or implementation tasks that can run without shared write state. Dispatch native subagents through the owning sp-* workflow.
compatibility
Requires spec-kit project structure with .specify/ directory

03 SKILL.md 正文

Dispatching Parallel Agents (Spec Kit Plus)

Use this skill when there are 2+ independent lanes: unrelated failing tests, separate bugs, distinct modules, read-only evidence lanes, testing-system build lanes, map scan/build lanes, or implementation tasks with non-overlapping write-sets.

The stable default is current-runtime dispatch. Do not turn parallel work into a user coordination problem when native subagents are available.

Current routing vocabulary:

  • Use parallel subagents only after the owning workflow selects delegation and confirms that parallelism materially improves the critical path or evidence quality.
  • Dispatch one-subagent when one safe lane is ready.
  • Dispatch parallel-subagents when two or more independent lanes can run concurrently.
  • If delegation becomes unsafe or unavailable, re-evaluate the owning workflow's route instead of treating parallelism as mandatory.
  • Do not use old strategy labels as routing choices.

When to Use

  • 2+ independent lanes can be understood or changed without waiting on each other.
  • Each lane has a clear owner, scope, and verification target.
  • Write-set overlap is absent or can be explicitly sequenced at a join point.
  • Running the lanes in parallel materially improves throughput, evidence quality, or verification confidence.

Process

  1. Route first: Select the owning sp-* workflow before dispatch. Common canonical routes are sp-quick, sp-debug, sp-map-scan, sp-map-build, and sp-implement. When telling the user what to type, use placeholders such as $sp-quick, $sp-debug, $sp-map-scan, $sp-map-build, or $sp-implement.
  2. Split lanes: Name each lane, purpose, read context, write-set, forbidden paths, shared surfaces, and verification target. sp-debug dispatches parallel agents only when the investigation exposes independent evidence lanes; small focused investigations may stay leader-inline under the debug session contract.
  3. Check conflicts: Do not dispatch two writers to the same file or shared state unless one lane is explicitly read-only or the workflow defines a safe join point.
  4. Packetize just in time: Compile validated packets only for the selected current lanes. Raw task text is not enough and future batches stay unexpanded.
  5. Dispatch native subagents: Use parallel-subagents on the native-subagents surface first, such as Codex spawn_agent/wait_agent, Claude Task, or the active CLI's equivalent. Keep the leader focused on integration and conflict resolution.
  6. Join on structured handoff: Each worker must report changed files, verification run, failures, open risks, and whether its acceptance target is met. Integrate only after the handoff is specific enough to review.

Fallbacks

  • Use sp-teams only when Codex work needs durable team state, result files, explicit join tracking, or lifecycle control beyond one in-session burst.
  • Use a separate terminal only when the current runtime has no native subagents and no managed-team path can safely represent the lanes.
  • If lanes share write state or one fix may resolve the others, keep the work in one sequential workflow run until the dependency is clarified.

Behavioral Rules

  • Do not fix unrelated systems in one leader-inline pass when current-runtime native subagents can handle independent bounded lanes.
  • Do not dispatch agents that will modify the same write-set concurrently.
  • Do not ask the user to manually coordinate parallel terminals as the first option.
  • Do not mark a lane complete from silence, idle status, or a vague summary.
  • If the first split reveals hidden coupling, stop parallel execution and move the coupled work behind one owner or join point.

Red Flags

  • "Fix all failing tests" with failures spread across unrelated modules.
  • Multiple implementation tasks touch different write-sets, but the leader is about to execute them serially without checking native subagents.
  • A worker prompt does not include a write-set, forbidden paths, or verification target.
  • A handoff says "done" without file paths and command output summary.
  • The fallback plan starts with "open another separate terminal" even though the current runtime supports native subagents.

04 同仓库更多 Skills

chenziyang110
chenziyang110/launchdeck

launchdeck-agent

Codex host adapter contract fixture for Launchdeck.

★ 0
chenziyang110
chenziyang110/launchdeck

launchdeck-agent

Claude host adapter contract fixture for Launchdeck.

★ 0
chenziyang110
chenziyang110/launchdeck

launchdeck-agent

Use this skill aggressively for local project lifecycle operations even when the user does not mention Launchdeck: start this project, run the dev server, dev, build, test, restart service, stop service, show logs, inspect ports, port is occupied, clean cache; and when the user explicitly asks to configure this local project for Launchdeck or create or generate a project-adapted `.launchdeck.yml`, including 为这个项目生成 Launchdeck 配置. Do not use for ordinary code edits, docs-only work, general explanations, API design discussion, or production deployment unless the user also asks to operate or configure a local project or service.

★ 0
chenziyang110
chenziyang110/launchdeck

zig-testing

Use ONLY when user explicitly requests Zig test code, built-in test runner configuration, test failure diagnosis, or coverage for Zig projects. Triggers: "write test zig", "zig test", "testing.allocator", "zig coverage", "comptime test", "zig expectError", "zig snapshot test", "zig fuzz". Do NOT activate for: general Zig coding without test context, build.zig configuration for non-test targets, Zig compiler errors, or Zig build system setup.

★ 0
chenziyang110
chenziyang110/launchdeck

webapp-testing

Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.

★ 0
chenziyang110
chenziyang110/launchdeck

verification-before-completion

Use when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always

★ 0