agentleFS
Cody-SimsCody-Sims/agent-skills/skills/parallel-worktree-delivery/SKILL.md

parallel-worktree-delivery

Skill0 starsChanged 4 months agoMITView on GitHub

Executes multi-agent Git worktree fleets with exclusive ownership, dependency waves, serialized integration, and merge gates. Use when delivering independent streams on separate branches/worktrees; not for task planning alone, one small change, or GitHub-hosted cloud agents.

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 4 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Run only an authorized fleet.

    line 13
  2. Very specific:

    Read path instructions; inspect branch/status/worktrees/operations and verify base/integration checkout.

    line 18
  3. Distinctive:

    Before delegation, complete the capability preflight in fleet setup: resolve required skills to supported native names or authorized source …

    line 25
  4. Distinctive:

    Read integration playbook before merges.

    line 56
  5. Distinctive:

    Run all repository-required per-merge gates after each merge.

    line 61

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Parallel Worktree Delivery

Run only an authorized fleet; loading is not consent to spawn, install, or mutate Git/remote state.

Prepare

  1. Read path instructions; inspect branch/status/worktrees/operations and verify base/integration checkout. Never stash, commit, switch, or overwrite others' live work.
  2. Bound streams/concurrency/time/cost to approved resources. Delegate substantial independent work only; honor requested runtime/model, invent no controls.
  3. Assign separate branches/worktrees and exclusive files; sequence dependencies in waves. Coordinator owns shared files; exact-section exceptions serialize writes.
  4. Before delegation, complete the capability preflight in fleet setup: resolve required skills to supported native names or authorized source paths and record actual messaging limits. Carry verified routes into worker prompts; block unsupported dependent work instead of installing globally or expanding permissions. Isolate or serialize writable dependencies, caches, outputs, databases, and ports. No concurrent shared installs; bound CPU-heavy jobs.

Dispatch and collect

  1. Supply objective/stop, branch/worktree/base, ownership, instructions/setup, scoped checks, allowed actions, and the capability contract. Workers confirm loading routes before dependent work and report cross-boundary needs, not edit.
  2. Record runtime/provider, requested and observed model (or unknown), usable worker handle, and owned repository/worktree/branch/base/tip in the handoff. Continue independent work. Use notifications/bounded known-ID waits, not tight polling/discovery or duplicate workers. Collect paths/commits, exact checks/results, conflicts, and boundary requests.
  3. Session/task IDs from history are not necessarily addressable worker IDs. Confirm the runtime's visible handle and message capability before contacting an external or resumed worker. On an unavailable handle, preserve its checkout ownership and use a durable handoff or report blocked; do not repeatedly send, spawn a replacement owner, or infer that its files are free to edit.
  4. Send only to explicit known message-capable handles. Deduplicate recipients and batch within the host-reported limit; do not hardcode a provider cap or use broad scopes to bypass it. One-shot workers are result-only unless the host explicitly supports follow-up. Unknown capabilities are blockers, not permission to probe by sending.

Integrate

  1. Read integration playbook before merges. Freeze tips; review ownership/full diffs. Serialize integration/shared Git administration in the clean coordinator checkout.
  2. Merge one stream at a time; reconcile both conflict intents. Stop on unresolved ownership/behavior disagreements or failed checks.
  3. Run all repository-required per-merge gates after each merge. Without a mandated full gate, check integrated changes/interactions. Never substitute worker/pre-merge evidence or defer required gates to the final wave.
  4. Coordinator owns integration fixes; rerun affected checks/required gates before another merge without resource races. Review final diff/required final gates; no unchanged scoped reruns solely for handoff. Stage explicit owned paths; remote writes/clean-worktree cleanup need consent.

Output

Return waves/ownership, worker IDs/integrated tips, per-merge checks/results, conflicts/fixes, and pending work/blockers. No transcript dump; distinguish local integration from remote delivery.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.