Launch Competing Agents in Git Worktrees
Launch parallel coding agents on one task in isolated Git worktrees, with matched briefs and verified run tracking.
Before & after
Same prompt, same agent — the only difference is the skill.
I can help you explore different authentication approaches. You could try a few different strategies and see which one works best for your use case. Some teams use OAuth, others use JWT, and some still use sessions. Each has pros and cons depending on your requirements. What kind of API are you building, and how many users do you expect? That would help me suggest the best approach to explore.
SPAWN CONFIGURATION: Auth Architecture Competition AGENT 1: Session-Based Auth (Baseline) • Isolated worktree: auth-session-baseline • Task: Implement Express middleware for HTTP-only cookie sessions; Redis-backed store; CSRF tokens on POST • Deliverable: Runnable server + integration test suite • Target metrics: Time-to-auth <50ms, session memory per user <200 bytes AGENT 2: JWT Stateless Auth (Challenge A) • Isolated worktree: auth-jwt-stateless • Task: Implement JWT issuance/verification; refresh token rotation; revocation via blacklist cache (Redis) • Deliverable: Same test suite (no framework changes, same API contracts) • Success criteria: Faster initial auth vs Agent 1, but measure refresh overhead AGENT 3: OAuth 2.0 Delegated (Challenge B) • Isolated worktree: auth-oauth-delegated • Task: Integrate with provider; handle scope negotiation, consent screen, callback verification • Deliverable: Drop-in SDK; same test pass rate as Agent 1/2 • Trade-off focus: User experience (consent flow friction) vs zero password storage EVALUATION CRITERIA (all agents benchmarked identically) ✓ Test suite pass rate (baseline: 98%+) ✓ Auth latency p95 (target: <75ms) ✓ Code complexity (cyclomatic complexity per method <8) ✓ Security audit: CVE exposure (0 known vulnerabilities) ✓ Scaling load: 1000 concurrent sessions (memory growth rate) JUDGMENT CALL No winner chosen by default. Each branch merges to feature branches; final choice depends on downstream requirements (mobile vs web, user base size, operational maturity).
About this skill
“Spawn three agents” should produce three independent attempts at the same task. This skill pins their starting commit, gives each agent a separate branch and worktree, and verifies that each agent actually starts in its assigned directory.
What it does
- Reads the initialized session's task, agent count, baseline, and constraints without inventing missing AgentHub commands or configuration.
- Creates distinct worktrees from one exact commit and records their branch and agent mappings.
- Sends identical task briefs with matching acceptance criteria, context, and resource limits.
- Checks isolation beyond source files, including ports, databases, generated output, and shared credentials.
- Tracks startup, completion, failures, and reported versus independently observed checks.
- Preserves every attempt for comparison and hands off branch locations without merging or discarding work.
What you get
A launched set of competing coding agents and a verified run ledger showing each agent's baseline, worktree, branch, status, and result location. If execution is unavailable, you get a clearly labeled launch plan and dispatch briefs with the exact blocker identified.
Best for
Developers using Claude Code or another agent-capable environment who already have a defined task and want several independent implementation attempts before choosing an approach.
SKILL.md preview
---
name: launch-competing-agents-in-git-worktrees
description: >-
Launch competing coding subagents for an initialized AgentHub session by pinning one starting commit, creating distinct Git worktrees, dispatching matched task briefs, and verifying agent-to-worktree assignments. Use when the user says "/hub:spawn", "spawn 3 agents", "launch competing agents", "try this task in parallel", "start the initialized AgentHub session", "my agents keep editing the same files", or "give every agent its own worktree".
---
# Launch Competing Agents in Git Worktrees
… (sign up to view the full skill)More development skills
View all Development skills →Performance Bottleneck Profiling: Node, Python, Go
Classifies a slowdown, gives exact profiler commands, reads the output, and proves the fix with same-conditions p95/p99 numbers.
Feature Scaffolding with Working Tests
Turn a short feature spec into code, runnable tests, and integration wiring that match your existing project.
Feature Flag Rollouts and Kill Switches
Design staged flag rollouts, verify rollback paths, and audit flag debt with explicit defaults, gates, and retirement plans.
Microcontroller Firmware and RTOS Engineering
Build and debug MCU firmware with bounded ISR/RTOS timing, safe DMA ownership, and measured power budgets.