mirror of
https://github.com/affaan-m/ECC.git
synced 2026-08-20 14:32:24 +02:00
someiyoshino-labandGitHub
4a4fa907f5
The 'Parallel Task Execution' rule encourages agents to spawn subagents, but defines no completion contract. Observed failure mode: subagents followed the rule, spawned their own children, and returned 'waiting for background agents' as their final answer. All children completed successfully, but their results were orphaned because a parent whose turn has ended cannot receive completion notifications - leaving zombie 'running' tasks and lost work. Adds three rules that apply at every delegation depth: 1. Your final message IS the deliverable (never end with 'waiting') 2. If you delegate, you own collection (no fire-and-forget) 3. Decompose only when work cannot fit in one context
2.5 KiB
2.5 KiB
Agent Orchestration
Available Agents
Located in ~/.claude/agents/:
| Agent | Purpose | When to Use |
|---|---|---|
| planner | Implementation planning | Complex features, refactoring |
| architect | System design | Architectural decisions |
| tdd-guide | Test-driven development | New features, bug fixes |
| code-reviewer | Code review | After writing code |
| security-reviewer | Security analysis | Before commits |
| build-error-resolver | Fix build errors | When build fails |
| e2e-runner | E2E testing | Critical user flows |
| refactor-cleaner | Dead code cleanup | Code maintenance |
| doc-updater | Documentation | Updating docs |
| rust-reviewer | Rust code review | Rust projects |
| harmonyos-app-resolver | HarmonyOS app development | HarmonyOS/ArkTS projects |
Immediate Agent Usage
No user prompt needed:
- Complex feature requests - Use planner agent
- Code just written/modified - Use code-reviewer agent
- Bug fix or new feature - Use tdd-guide agent
- Architectural decision - Use architect agent
Parallel Task Execution
ALWAYS use parallel Task execution for independent operations:
# GOOD: Parallel execution
Launch 3 agents in parallel:
1. Agent 1: Security analysis of auth module
2. Agent 2: Performance review of cache system
3. Agent 3: Type checking of utilities
# BAD: Sequential when unnecessary
First agent 1, then agent 2, then agent 3
Delegation Completion Contract
Applies to every agent at every depth (parent, child, grandchild):
- Your final message IS the deliverable. Never end your turn with "waiting for background agents" — a spawned task is not a completed task. Ending your turn while children are running orphans their results (completed children cannot notify a parent whose turn has ended).
- If you delegate, you own collection. Wait for results, integrate them, then return. Fire-and-forget delegation is forbidden.
- Decompose only when the work cannot fit in one context. Do not re-delegate a task already sized for a single agent — depth is an outcome, not a plan.
Rationale: observed failure mode — research agents followed "Parallel Task Execution" above, spawned children, and returned "waiting" as their final answer. All children completed successfully but their results were orphaned. The parallel rule without a completion contract produces zombie tasks.
Multi-Perspective Analysis
For complex problems, use split role sub-agents:
- Factual reviewer
- Senior engineer
- Security expert
- Consistency reviewer
- Redundancy checker