wendy-iterating — Claude Code skill: fix loop with PRs
Runs a find-fix-ship loop: each change gets its own worktree, a subagent carries it to green CI, and the loop stops itself when nothing is left.
npx skills add wendylabsinc/claude-skills --skill wendy-iterating -a claude-codeWhat wendy-iterating does
The skill handles the plumbing of the loop: keeping fixes apart in worktrees, handing them to a subagent, driving pull requests, spacing out checks and shutting down. It has no idea how your project is tested. That part comes from a companion skill such as /wendy-cloud-testing, which tells it what command to run, where runtime errors show up, what counts as clean, and which failures are already known. Without that companion the loop has nothing to measure.
Every change follows the same path. The skill reads the current branch and keeps it as the base. Then it adds a worktree under $BASE/.worktrees/$BRANCH on a branch named fix/<short-description>. A subagent gets the files and lines, the root cause and the fix, runs the targeted test inside that worktree and commits. Next come gh pr create and gh pr checks --watch; nothing moves on until the checks pass. Merging is gh pr merge --squash --delete-branch, after which the worktree is removed. Commits never go straight onto the current branch.
Some fixes should not be made inline. If the change runs past 30 lines, crosses subsystems, or the cause is still murky after 15 minutes of digging, or architecture has to shift, the skill points you to a Linear issue instead. The wait before the next check follows the outcome: 5 minutes when something was found and fixed, 10 when a bug was found but left alone, 30 when everything is clean. Three clean rounds in a row end the loop, and any bug resets that counter.
What it helps with
- run a continuous bug-finding and fixing loop
- push a feature until every test passes
- keep each fix in its own worktree away from the main checkout
- route every change through a pull request and wait for CI
- let the loop stop on its own once nothing is broken
Example prompts for the agent
- Iterate on this feature with
/wendy-iterating, and use/wendy-cloud-testingto decide how to test for bugs - Start the fix loop with
no-auto-terminateso it keeps running - Run another iteration: tests, logs, then the skill's fix workflow
How to install wendy-iterating
1Skills CLI (Vercel)
npx skills add wendylabsinc/claude-skills --skill wendy-iterating -a claude-code2Manually: put the files in place
~/.claude/skills/wendy-iterating/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/wendy-iterating/
1Skills CLI (Vercel)
npx skills add wendylabsinc/claude-skills --skill wendy-iterating -a codex2Manually: put the files in place
.agents/skills/wendy-iterating/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add wendylabsinc/claude-skills --skill wendy-iterating -a cursor2Manually: put the files in place
.cursor/skills/wendy-iterating/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install wendylabsinc/claude-skills wendy-iterating2Manually: put the files in place
.github/skills/wendy-iterating/SKILL.md1Skills CLI (Vercel)
npx skills add wendylabsinc/claude-skills --skill wendy-iterating -a gemini-cli2Platform CLI
gemini skills install https://github.com/wendylabsinc/claude-skills --path wendy-iteratingManually: put the files in place
.devin/skills/wendy-iterating/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add wendylabsinc/claude-skills --skill wendy-iterating -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about wendy-iterating
How do I install wendy-iterating in Claude Code?
Run npx skills add wendylabsinc/claude-skills --skill wendy-iterating -a claude-code. The skill lands in ~/.claude/skills/wendy-iterating/SKILL.md.
What does wendy-iterating do?
It drives a find-fix-ship loop: isolates each fix in a worktree, dispatches a subagent, opens a pull request, waits for CI and merges the branch. It also paces the checks and ends the loop after 3 clean rounds.
Does it need a separate testing skill?
Yes. wendy-iterating cannot test your project by itself. Pair it with something like /wendy-cloud-testing, which supplies the test command, the log check, the clean definition and known failure modes.
