Code Forge — Parallel

⚡ Execution Entry Point (READ THIS FIRST)

When this skill is loaded, you MUST immediately begin executing the Workflow below — do not wait, do not summarize, do not ask "what should I do now". Skills are operational manuals, not reference documents. Read the first executable step, perform it, then the next, etc., until the workflow completes or you reach an AskUserQuestion checkpoint.

If the harness shows you Successfully loaded skill · N tools allowed, that message means the SKILL.md content was injected into your context — it does NOT mean the skill has run. Skills do not "run" autonomously; you run them by executing the Detailed Steps below.

If you find yourself about to say "the skill didn't produce output", "skill 仍未输出", "falling back to manual handling", "回退到手动处理", or anything similar, STOP. You have misunderstood how skills work. Go directly to the first executable step and start.

The first user-visible action of this skill should be either (a) the output of the first step, or (b) an AskUserQuestion if the first step needs disambiguation. Never an apology, never a fallback, never silence.

---

Dispatch independent sub-agents to solve multiple unrelated problems concurrently.

When to Use

• 3+ test files failing with different root causes
• Multiple independent subsystems broken simultaneously
• Several unrelated bugs reported at once
• Any situation with 2+ problems that share no state

When NOT to Use

• Failures are related (fix one, others may resolve)
• Problems share state or dependencies
• Need full system state to diagnose (single-threaded investigation needed)
• Only 1 problem exists

For task execution within a feature: Use /code-forge:impl which has built-in parallel task support.

Core Principle

One agent per independent problem domain. Let them work concurrently.

Workflow

Step 1: Identify Problems

List all problems. For each, note:

• What is failing (test file, error message, component)
• Which files/modules are involved
• Whether it shares state with other problems

Step 2: Assess Independence

Build a dependency matrix:

Problem A ←→ Problem B: independent? (yes/no, why)
Problem A ←→ Problem C: independent? (yes/no, why)
Problem B ←→ Problem C: independent? (yes/no, why)

Group dependent problems together. Each independent group gets its own agent.

Step 3: Dispatch Agents

For each independent problem group, launch a Agent(subagent_type="general-purpose"):

Agent prompt structure:

You are investigating and fixing: {problem description}

## Scope
- Files involved: {list}
- Error messages: {paste}
- Expected behavior: {description}

## Constraints
- Only modify files in your scope
- Use code-forge:tdd methodology (write failing test first)
- Use code-forge:debug methodology (root cause before fix)

## Output Required
- Root cause (1-2 sentences)
- Files changed (list)
- Test results (pass/fail counts)
- Summary of fix (1-2 sentences)

CRITICAL: Launch all agents in a single message using multiple Agent tool calls. This enables true parallel execution.

Step 4: Review and Integrate

After all agents complete:

1. Check for conflicts — did any agents modify the same file?
   • No conflicts: proceed
   • Conflicts: resolve manually, prefer the more targeted change
2. Run full test suite — verify all fixes work together
3. Report results:

Parallel dispatch complete: {N} agents, {M} problems resolved

Agent 1: {problem} → {root cause} → {status}
Agent 2: {problem} → {root cause} → {status}
Agent 3: {problem} → {root cause} → {status}

Full test suite: {pass}/{total} passing

Example

3 test files failing: auth.test.ts, payment.test.ts, email.test.ts

Independence check:
  auth ←→ payment: independent (different modules, no shared state)
  auth ←→ email:   independent (different modules)
  payment ←→ email: independent

Dispatch 3 agents in a single message:
  Agent 1: "Fix auth.test.ts — files: src/auth/, error: missing token validation"
  Agent 2: "Fix payment.test.ts — files: src/payment/, error: decimal precision"
  Agent 3: "Fix email.test.ts — files: src/email/, error: template not found"

Results: 3/3 resolved, full suite 128/128 passing

Common Mistakes

• Dispatching agents for related problems (they'll step on each other's changes)
• Not providing enough context in agent prompts (agent wastes time re-discovering)
• Launching agents sequentially instead of in a single parallel message
• Skipping the full test suite after integration
• Dispatching too many agents (>5) — diminishing returns, harder to integrate