agentsclimarketplace

Subagent dispatch

Skill IrfanSadiqRahat/constellation/methodology/subagent-dispatch

Dispatch fresh subagents per task with typed artifact handoff and two-stage review.From its SKILL.md

Install
npx -y skills add IrfanSadiqRahat/constellation --skill subagent-dispatch

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

2.4 KB, 542 tokens by cl100k_base, as published. Nobody here has run it

subagent-dispatch

When activated

A Plan exists with N tasks. Tasks are independent enough to dispatch in parallel OR sequential with strict handoff.

Dispatch model

For each task T in the plan:

  1. Materials briefing — gather only what T needs:
    • The task spec (T block from the Plan)
    • The input artifact (T.consumes)
    • The role's SKILL.md (agents/<role>/SKILL.md)
    • Relevant prior decisions (links, not bodies)
  2. Dispatch — spawn the subagent with only the briefing, no extra context
  3. Output — subagent returns the artifact (T.produces) shaped per schema
  4. Review stage 1: spec compliance — does output match the declared shape?
  5. Review stage 2: quality — does it pass acceptance criteria?
  6. Merge — write artifact to the team workspace; mark task done

Parallel vs sequential

  • Sequential when phase N+1 consumes phase N's output (default for pipelines)
  • Parallel when tasks share input but produce independent outputs (e.g. design-lead and ux-researcher both consume PRD)
  • Fan-out + merge when N agents tackle the same problem and a coordinator picks the best (used for adversarial review)

Two-stage review (required)

stagechecksrun by
1: specshape of output, presence of required fields, type of in/out matchesmechanical (schema validation)
2: qualityacceptance criteria, smell-checks from role's SKILL.md, no banned patternsanother subagent (peer role)

If either review fails, dispatch again with the failure as additional briefing. Three failures = escalate to human.

Materials hygiene (the moat)

  • Fresh subagent for each task. No accumulated context drift.
  • Briefing is minimal. Less is more reliable.
  • No tool access not needed for this task. Reduces blast radius.
  • Output schema enforced. Free-form output is the enemy.

Why this skill is opinionated

  • Typed artifact contracts — review stage 1 is mechanical, not subjective
  • Fan-out + merge pattern — multiple agents same problem, vote-merge
  • Role's SKILL.md is materials — briefing includes the role's playbook
  • Two-stage review by a peer role — security-architect reviews appsec-engineer, etc.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 326,144. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.