parallel-fix
Skill9 starsChanged 5 months agoMITView on GitHub
Use to FIX or implement a LIST of mostly-independent changes faster by doing them concurrently instead of one-at-a-time. Triggers on "并行修复 / 并行修这份清单 / 批量修复 / 把审计结果(清单)修了 / 并行开发 / 一起改", and "parallel fix / fix this audit list in parallel / apply these findings / batch-fix / implement these tasks concurrently". Best right after a business-defect-audit or any review that produced a findings list. NOT for a single change (just do it directly), and NOT a license to blindly parallelize tightly-coupled work — high-risk concurrency/resume/budget/determinism fixes are flagged for careful single-fix TDD instead.
In this repository3 other files it gives its agents.
- AGENTS.mdAGENTS.md
- business-defect-auditSkill
- dynamic-workflowSkill
The file
Files
Parallel Fix
Speed up the fix/implement half of the loop. A normal agent fixes a list serially (explore → fix → test → commit, repeat) — slow. This skill generates every fix concurrently, then integrates + verifies once.
Engine / backend / sandbox / journal mechanics: reuse the dynamic-workflow skill. This skill carries only the method.
The key idea (and its honest limit)
The slow, parallelizable part is "analyze each finding + write the correct patch" — independent per finding. So fixers run in parallel but READ-ONLY, each returning a unified diff (not editing the tree — parallel writers to one repo race and the engine will reject them). A single serial step then applies the diffs, resolves conflicts, and runs the full suite once.
Honest limits — say these to the user, don't hide them:
- Integration + verify is serial (one suite run on the merged result) — that's the bottleneck you can't parallelize away.
- Overlapping fixes (same file/lines) can't be blindly merged — the apply step puts conflicts in a
conflictedlist for manual handling. - Parallel fixers don't see each other, so a fix may assume code another changed — the single full-suite verify catches breakage; a reconciliation pass may be needed.
- Realistic speedup ≈ 2–4× (more independent findings → more speedup), NOT N×.
Steps
Generate .codex-flow/generated/<slug>.fix.workflow.ts from references/fix-template.workflow.ts (import-free, plain JSON Schema), set FINDINGS to the real list (from the audit / the user), then codex-flow run it. The phases:
- Propose (
parallel,sandbox: "read-only") — one agent per finding. Each reads the code and returns the smallest correct unified diff + ariskrating. Read-only ⇒ safe to share onecwd. Routerisk: "high"items (concurrency / resume / budget / determinism) out of the blind-parallel path — recommend the user fix those with single careful TDD (a reproducing test first), because that's where parallel-blind fixes regress. - Apply + verify (
phase, single agent,sandbox: "workspace-write",cwd: process.cwd()) — apply the diffs in a conflict-minimizing order; record un-mergeable ones underconflicted; run the project's full test + typecheck once; reporttestPassedhonestly.
Rules
- Fixers are read-only and only propose diffs — they never edit files in parallel.
- Exactly one writable apply step, run alone (no concurrent writer on the same
cwd). - Never claim a fix landed without the suite passing. Report
applied/conflicted/skippedandtestPassedtruthfully. - High-risk fixes → recommend single-fix TDD, don't fold them into the blind batch.
- Summarize for the user: what landed, what conflicts need manual work, whether tests pass, and the journal path.
Full method (grouping, diff format, conflict handling, when to fall back to serial): references/fix-method.md.
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.

