Batch Worker: 100 AI Agents Running in Parallel
The Problem
Auditing a codebase takes hours when you go file by file. Content creation, search, fixes — every task is bottlenecked by sequential execution.
The Solution
Batch Worker is an OpenClaw skill that dispatches up to 100 AI agents in parallel with staggered launch to avoid rate limits.
Three-Step Pipeline
1. ai_planner -> Analyze project, generate audit plan with 100 soldier prompts
2. core_taskPipeline -> Dispatch 100 agents in staggered batches (10/batch x 20ms)
3. ai_collector -> Collect reports, deduplicate, rank by severity — zero LLM tokens
104 Audit Dimensions
| Domain | Dimensions |
|---|---|
| Security | 42 (injection, XSS, CSRF, auth, encryption...) |
| Architecture | 12 (boundaries, cycles, idempotency, resilience) |
| Performance | 12 (N+1, caching, memory leaks, lock contention) |
| Code Quality | 10 (complexity, error handling, dead code) |
| Language-specific | 10 (Promise, async, EventEmitter, Stream) |
| DevOps | 8 (observability, CI/CD, config, containers) |
| Compliance | 4 (data privacy, audit trail, accessibility) |
83 Task Types
From code audit to content creation, search to fix, translation to analysis — batch-worker handles them all.
Zero-Token Cleanup
After 100 agents finish, ai_collector automatically extracts JSON findings, deduplicates, and merges — using zero LLM tokens. Pure script, no hallucination.
GitHub: https://github.com/haoyun18881-beep/batch-worker
Docs: https://haoyun18881-beep.github.io/batch-worker/