Reviews a plan or shell script that lets several coding agents work in parallel, each in its own git worktree, and lists only the five mistakes that make the agents collide, one line each with a fixed code, the place, why it breaks and the fix - or exactly No findings. It catches one branch checked out in two worktrees or forced onto a second one, two agents started on the same fixed port, an untracked secret file missing in a new worktree or written by two agents, a build folder or database file written by more than one agent, and worktrees that are never removed. It leaves out the safe look-alikes, such as two worktrees on different branches cut from one commit or a shared folder that every agent only reads. Use before launching parallel agents, to audit an orchestration script, or to debug agents that overwrite each other.
Parallel Agent Worktree Review is a tested SKILL.md that reviews a plan or shell script that lets several coding agents work in parallel, each in its own git worktree, and lists only the five mistakes that make the agents collide, one line each with a fixed code, the place, why it breaks and the fix - or exactly No findings; an agent buys it once for $0.01 over x402.
Not for
How an agent stops or retries, what the agents write, locking worktrees, sparse checkout and bare repositories. It reads one pasted plan or script and judges only what is written in it.
Tested, honestly
Tested 2026-10-19 with a strong and a weak model.
- SonnetStrong model, claude-sonnet-5-5
- Right on all 3 cases, read by hand. On the script with a forced add it named the second worktree as the branch problem and the missing removal at the end, and left out the read-only fixtures folder and the separate branch cut from the same commit. On the three-bugfix plan it gave exactly the shared port and the shared database file and left out the copied env file and the proper cleanup. On the tidy plan it answered exactly No findings.
- HaikuWeak model, claude-haiku-5-5
- Right on all 3 cases, read by hand. It gave the same two lines as the strong model on each planted plan, with sound fixes, and answered exactly No findings on the tidy plan. It spent far more output tokens than needed on the clean plan, but the final answer was one sentence.
Full test summary
Example
Our own test text, before and after the skill ran. Excerpts only.
English · claude-sonnet-5-5
Before
#!/bin/bash
# Run three agents in parallel on the search feature
set -e
cd ~/projects/shop
git worktree add ../shop-a feature/search
git worktree add -f ../shop-b feature/search
git worktree add -b agent/c ../shop-c HEAD
chmod -R a-w /data/fixtures
(cd ../shop-a && PORT=4001 run-agent --task search-ui --fixtures /data/fixtures) &…
After
BRANCH-TWICE — worktree ../shop-b (`git worktree add -f ../shop-b feature/search`) — feature/search is already checked out in ../shop-a and the add is forced past git's refusal, so both agents work on one branch — give each agent its own branch, e.g. `git worktree add -b agent/b ../shop-b feature/search`…
What is in the file
- What git guarantees, as published in the git documentation on 19 October 2026
- The answer
- The codes
- Rules
- Work in this order
- Short example
Languages
English. Tried in: English.
License
Perpetual, non-exclusive; use and modify for yourself incl. paid work; no resale or republishing. Holder: Georgi Kalchev, aiskills402.com. Full terms.
Versions
Current version 1.0.0, updated 2026-10-19. Whoever bought an earlier version gets new ones free through the same re-download token.
v1.0.0 · 2026-10-19
First release, written in the light-test batch of 19 October 2026 (three cases, price $0.01). The idea is ours; no foreign skill was opened, facts come from public documentation (see notes/ideas.md). Output: see the skill description.
Test: three cases written by hand (two planted, each with a safe look-alike, one clean control), one run on Claude Sonnet and one on Claude Haiku, no run without the skill, no iteration on the skill text after the runs. `test/control.mjs` shows that the checks pass on the ideal answers and fail on realistic wrong ones. See `test-summary.json` for the verdicts read by hand.
FAQ
Shared starting commit for every worktree: a collision?
No. Two worktrees on different branches that start from one commit are normal and git allows them. The skill reports one branch held by two worktrees, or an add that is forced past git refusal.
Is a shared folder that agents only read a problem?
No. A fixtures folder made read-only and read by every agent is left alone. The skill reports a shared path that more than one agent writes, such as one database file in a temp folder or a common build folder.
Does it warn about missing env files?
Yes, when the plan never copies or links them. A fresh checkout holds only committed content, so an untracked env file or key is absent until the plan puts it there. The git manual does not state this in so many words, and the skill says as much in its text.
What was in the test?
Three plans were each read by both models in a single run: a script that forces one branch into two folders and never removes them, a plan that starts three servers on one port against one shared database file, and a tidy plan with ports computed per agent. On reading each answer, both models met all expectations.