github-orchestration

v2026.09.25

This skill should be used when coordinating multi-agent GitHub delivery work — dispatching teams, holding a blackboard, enforcing merge gates, and deciding what to run in parallel versus in sequence.

GitHub
安装命令
npx skhub add thelobbi/github-orchestration
Markdown
SKILL.md

GitHub Orchestration

Coordination patterns for multi-agent GitHub delivery.

Choosing a coordination pattern

PatternUse whenCost
SequentialEach step needs the previous step's outputSum of steps
Parallel + barrierYou need all results together — dedup across findings, an early exit on zeroSlowest step
Pipeline (no barrier)Each item flows through all stages independentlySlowest single chain
HierarchicalA coordinator delegates and synthesizesSlowest worker + synthesis
Adaptive loopRepeat until a convergence condition (drive-to-green)Unbounded — always cap rounds

Default to pipeline over barrier. A barrier is justified only when a stage genuinely needs cross-item context. "I need to flatten the array first" is not a justification — do the transform inside a stage.

What parallelizes in GitHub work

Independent, run concurrently:

  • Review lenses over the same diff
  • Security scans across alert classes
  • Metric collection across dimensions
  • Per-PR triage across open PRs

Genuinely sequential:

  • Branch → implement → PR (each needs the last)
  • Landing a stack (each landing changes the next base)
  • Migration → code → cleanup

The blackboard

Agents write findings to a shared structure; the coordinator reads and decides.

  • Contradiction is information. Two agents disagreeing means one has context the other lacks. Investigate; do not average their confidence.
  • Deduplicate by root cause, not by text. Two lenses describing the same defect differently is one finding.
  • Record what was dropped and why. A board that reports only survivors is indistinguishable from a board that found nothing.

Delegation

  • Narrow mandate plus an explicit return contract. "Review this PR" produces mush; "find correctness defects, return file/line/failure-scenario" produces findings.
  • Keep subagent prompts under ~400 words — subagents inherit a large context chain before your prompt, and long prompts raise rejection rates.
  • Prefer a named specialist over a general-purpose agent.
  • Never spawn duplicate agents hoping one succeeds. Spawn them with different lenses so disagreement carries signal.

Merge gates

These are absolute. An agent may not route around them:

  1. Every required check green at the current head sha
  2. Review quorum met, no outstanding CHANGES_REQUESTED
  3. No unresolved threads when protection requires resolution
  4. Clean mergeability against the base
  5. No branch-protection bypass or admin override
  6. Human approval for humanReviewPaths

Re-verify against live state immediately before merging. A gate that passed ten minutes ago is not evidence about now.

Reporting

Report per phase, with a marker per phase, and never claim a phase you skipped. A summary that reads complete while a step was silently dropped is the worst possible output — it converts a partial result into a false one.

See also

  • merge-queue — queue and protection semantics
  • review-protocols — the adversarial board
  • ../workflows/README.md — declarative workflow authoring
发现
标签

此技能尚未发布标签。

版本
最新版本元数据

版本

v2026.09.25

发布时间

2026年9月25日

分类

未分类

许可证

MIT

源路径

plugins/delivery-orchestrator/skills/github-orchestration

默认分支

main

最新提交

2f1269c

Tree SHA

629e050