github-swarm-orchestration

Coordinates multi-agent research work through GitHub issues, branches, and PRs with explicit review and verifier gates. Use when running a swarm, parallel issues, orchestration log, feat/issue-N branches, merge order, consistency pass before merge, or Day 3-style lab workflow.

Jump to install

Repository: meleantonio/bac-starter
Last source activity: April 8, 2026 at 21:22
Detected SKILL.md language: English
Stars: 1
Forks: 1

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

Back to search Back to installation

name	github-swarm-orchestration
description	Coordinates multi-agent research work through GitHub issues, branches, and PRs with explicit review and verifier gates. Use when running a swarm, parallel issues, orchestration log, feat/issue-N branches, merge order, consistency pass before merge, or Day 3-style lab workflow.

GitHub Swarm Orchestration

When to use

User mentions: swarm, parallel agents, orchestration log, one issue one branch, Cloud Agent per issue, merge order, or materials/day3 workflow.
You need a repeatable captain loop: plan → issues → branches → PRs → review → verify → merge → follow-on issues.

Preconditions

If work is spec-driven: spec/tasks.md should be approved (use sdd-orchestrator / SDD skill first).
Git remote exists; gh auth status succeeds when using GitHub CLI.

Workflow

1. Choose spine

Spec-first: Read spec/tasks.md and map checkboxes to GitHub issues (delegate drafting to spec-tasks-to-github skill for issue bodies).
Issue-first: Use .github/ISSUE_TEMPLATE/swarm-task.md fields directly.

2. Build the issue DAG

Label independent work parallel.
Label ordered work sequential and add blocked-by:<N> for the prerequisite issue number.
Never start a worker on a blocked issue until the blocker is merged.

3. Initialize or update the log

Open notes/orchestration_log.md.
Add one row per active stream/issue: branch, issue, status, PR (when opened), reviewer.

4. Dispatch workers

Branch: feat/issue-<N>-short-slug unless the milestone declares rig/<stream> (then log it).
Optional creative worker name for commits and PR description (per project convention).
Use handoff files notes/handoff_<from>_to_<to>.md when streams consume each other's outputs.

5. Review gate

Invoke pr-reviewer subagent on each PR (readonly review first; then apply fixes on the branch).
If labels include domain:*, invoke the matching domain reviewer agent.

6. Verifier gate

replication-verifier: when replication paths, deps, or run instructions change.
artifact-coherence-auditor: when paper text, tables/figures, and replication must agree.

7. Merge gate

integration-coordinator (or human): enforce dependency order, rebase if needed, resolve conflicts.
Automerge only if issue has automerge-ok and CI + review requirements are satisfied.

8. Follow-on work

Verifier blockers → open new issues linked from the audit; do not silently expand an already-approved PR scope.

Guardrails

Follow .cursor/rules/research-reproducibility.mdc: relative paths, traceable outputs.
Keep notes/orchestration_log.md the durable state mirror alongside GitHub.
See reference.md in this folder for labels, gh snippets, and phase detail.

Output

At end of each captain iteration, report:

Updated issue/PR links
Current notes/orchestration_log.md row states
Pending blockers and next merge order