parallel-orchestration
Skill49 starsChanged 38 days agoMITView on GitHub
Template for multi-agent work in this repository where a main session coordinates parallel sub-agents and a single commit agent. Use only when the user explicitly asks for parallel or multi-agent orchestration.
- Installs packages
- Commits and pushes
1 other project uses a version of this.
Adopted by
Most recently taken up by
- PMDevSolutions/Aureliustheir own version, first seen Nov 9, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- PMDevSolutions/Aurelius★ 9same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository3 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- contract-reviewSkill
- liferay-debugSkill
The file
We split this skill into the rules it gives an agent
10 rules. About 0 are common, 5 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 19
If two sub-agents would touch the same file, serialize them — never let two agents race the same path.
- Very specific: line 25
Commit messages are one line.
- Very specific: line 30
When a sub-agent's remaining context budget drops to 5% or below, run /compact before continuing.
- Very specific: line 49
When merging a base branch (e.g.
- Distinctive: line 53
Use when the feature branch is authoritative and base-branch changes should be discarded in the conflicting files:
- Very specific: line 58
Confirm with the user which side is authoritative: before running any --ours / --theirs — these flags are destructive and silently drop the …
- Distinctive: line 62
Run verification commands (see I24).
- Very specific: line 74
Run yarn install / npm install / pnpm install to rebuild lock-file consistency against the resolved package.json.
- Distinctive: line 76
Only then stage, commit, and push.
- Distinctive: line 93
All merge-conflict commits still flow through the single commit agent — sub-agents never resolve conflicts and commit on their own.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Parallel Orchestration Workflow
Template for running multi-agent work where the main Claude Code session coordinates and sub-agents execute.
Roles
- Main agent — orchestration only. MUST NOT read, edit, or run code directly. Its job is to decompose the task, dispatch sub-agents, aggregate their reports, and decide the next phase.
- Sub-agents — execute one scoped unit of work each. Run in parallel whenever their file clusters are independent.
- Commit agent — a dedicated, single, serial agent responsible for every git commit. No other agent commits.
Parallelism rules
- Dispatch sub-agents in parallel whenever their working sets do not overlap. Partition by file cluster, not by task phase.
- If two sub-agents would touch the same file, serialize them — never let two agents race the same path.
- The main agent waits for all parallel sub-agents in a wave to report before starting the next wave.
Commit rules
- Commits are serialized through the single commit agent. Sub-agents stage nothing and push nothing.
- Commit messages are one line. No
Co-Authored-Bytrailers, no body text, no emoji. - One logical change per commit. If a sub-agent produced two unrelated fixes, the commit agent splits them.
Context hygiene
- When a sub-agent's remaining context budget drops to 5% or below, run
/compactbefore continuing. Preserve: discovered solutions, final decisions, file paths that matter. Drop: debug transcripts, failed attempts, raw tool output. - The main agent's own context should stay lean — it holds plans and summaries, not source code.
Four-phase pattern
- Explore — one or more read-only sub-agents map the relevant code, answer "what exists, where, and how is it wired?". Output: a written summary with absolute file paths.
- Design — main agent (or a dedicated design sub-agent) turns the exploration into a plan: file clusters, sub-agent assignments, dependencies. Output: a task list with clear ownership boundaries.
- Implement — parallel sub-agents edit their clusters. Each reports completion with a diff summary.
- Review → Simplify → Verify — a review sub-agent reads the diffs, a simplify sub-agent removes dead code / redundancy, and a verify sub-agent runs the actual build/test commands. Only after verify is green does the commit agent run.
Anti-patterns
- Main agent opens files "just to double-check" — forbidden; dispatch a sub-agent.
- Multiple agents committing in the same session — forbidden; the commit agent is the only writer of git history.
- Skipping the verify step because "the diff looks right" — forbidden; green build is the gate.
- Sub-agents spawning their own sub-agents — forbidden; only the main agent dispatches.
Merge Conflict Resolution
When merging a base branch (e.g. origin/master) into a feature branch produces conflicts, follow a deterministic pattern instead of hand-editing conflict markers.
Pattern I23 — feature branch wins
Use when the feature branch is authoritative and base-branch changes should be discarded in the conflicting files:
git fetch origingit merge origin/<base-branch>- Identify conflicting files from
git status. - Confirm with the user which side is authoritative before running any
--ours/--theirs— these flags are destructive and silently drop the other side's changes. git checkout --ours <file>(or--theirs <file>) for each conflicting path.git add <file>git commit --no-edit- Run verification commands (see I24).
git push <branch>
Pattern I24 — always re-run tests after a merge
Lock files (yarn.lock, package-lock.json, pnpm-lock.yaml) can auto-merge cleanly yet produce a broken dependency graph. Never trust a green merge — re-run the full unit and integration test suites before pushing.
Lock-file + manifest triad
package.json and its lock file almost always conflict together. The canonical recovery is:
- Resolve all three files (
package.json,yarn.lock/package-lock.json) with the same side (--oursor--theirs) — never mix. - Run
yarn install/npm install/pnpm installto rebuild lock-file consistency against the resolvedpackage.json. - Re-run
yarn test(unit + integration) to catch regressions from transitive dependency drift. - Only then stage, commit, and push.
Step template
1. git fetch origin
2. git merge origin/<base-branch>
3. List conflicting files
4. Confirm with the user which branch is authoritative
5. git checkout --ours <file> (or --theirs)
6. git add <file>
7. yarn install / pnpm install to re-sync lock files
8. yarn test / unit test / integration test to catch regressions
9. git commit --no-edit
10. git push <branch>All merge-conflict commits still flow through the single commit agent — sub-agents never resolve conflicts and commit on their own.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

