Skip to content
Skills Library
AI & agents · Planning

github-orchestration

/github-orchestration

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.

Repository
claude ↗
Licence
MIT
Last repository change
22.10.2026

Assessment and security

90/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content25 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
1
Scripts
0
HTML comments
0
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 22.10.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What it does and when to use it

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.

Original author description

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.

How it works

  1. Choosing a coordination pattern
  2. What parallelizes in GitHub work
  3. The blackboard
  4. Delegation
  5. Merge gates
  6. Reporting
  7. See also
Original SKILL.md headings · 80 lines, 496 words
SKILL.md preview
---
name: github-orchestration
description: 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.
version: 1.0.0
trigger_phrases: [orchestrate github, coordinate agents, delivery loop, merge gate, dispatch team]
categories: [github, orchestration, multi-agent, delivery]
author: github-orchestrator
created: 2026-08-11
updated: 2026-08-11
---

# GitHub Orchestration

Coordination patterns for multi-agent GitHub delivery.

## Choosing a coordination pattern

| Pattern | Use when | Cost |
| --- | --- | --- |
| **Sequential** | Each step needs the previous step's output | Sum of steps |
| **Parallel + barrier** | You need *all* results together — dedup across findings, an early exit on zero | Slowest step |
| **Pipeline (no barrier)** | Each item flows through all stages independently | Slowest single chain |
| **Hierarchical** | A coordinator delegates and synthesizes | Slowest worker + synthesis |
| **Adaptive loop** | Repeat 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

Versions

The database has no other skills named github-orchestration.

Database snapshot: 13.11.2026 · package version bb3dd84ba929 (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 100team-agent-orchestration
    Run team-based orchestration for agent squads using work items, ownership, agent Kanban, merge gates, and control pane handoffs. Use when coordinating an agent squad with work items, ownership, Kanban, and merge gates.
  • 100team-agent-orchestration
    使用 work items、ownership、agent Kanban、merge gates 和 control pane handoffs 运行团队式智能体编排。
  • 100team-builder
    Interactive agent picker for composing and dispatching parallel teams. Use when composing and dispatching a parallel team of agents for a task.
  • 96multi-profile-project-orchestration
    Use when coordinating multi-profile project work.
  • 100claude-code-agent-teams
    This skill should be used when the user asks to "create an agent team", "run a multi-agent team", "coordinate multiple Claude sessions", "use agent teams", or wants to parallelize work across multiple Claude Code instances.
  • 100team-lead
    (forwward) Orchestrates multi-agent engineering and business teams by analyzing tasks, selecting team shapes, assigning specialist roles, and coordinating parallel agent work with skill dispatch. Triggers on tasks needing 2+ agents, feature dev with frontend + backend, quality…

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.