Dispatching Parallel Agents

Register: TECHNIQUE Goal: Solve multiple independent problems in parallel time without context contamination between agents. Constraints: True independence required — no shared files, no sequential dependencies. Each agent gets its own clean context. pattern-globalize after any agent finds a bug. Adapt: number of agents, context depth, and coordination protocol to task complexity and available token budget.

When to Use

Use when ALL are true:

• 2+ independent failures or tasks (different subsystems, different test files)
• Each can be understood without context from the others
• Agents won't touch the same files

Don't use when:

• Failures might be related (fixing one may fix others — investigate first)
• Need full system context to understand the problem
• Agents would touch shared files

The Pattern

1. Identify Independent Domains

Goal: group failures by structural independence, not by file or feature proximity.

Domain A: auth module — 3 test failures
Domain B: data pipeline — 2 test failures
Domain C: API handlers — 1 test failure

Each domain independent: auth failure won't affect pipeline behavior.

2. Construct Focused Agent Context

Goal: give each agent exactly what it needs — no more, no less. Constraints:

• Never let subagents inherit your session context — they get confused
• Specify scope explicitly: "Do NOT change files outside tests/auth/"
• Specify output format: "Return: root cause summary + list of file:line changes"

Fix the N failing tests in <path>:

1. "<test name>" — <expected behavior from test>
2. "<test name>" — <expected behavior from test>

Your task:
1. Read the test file — understand what each test verifies
2. Find root cause using systematic-debugging Phase 1-3
3. Fix using tdd-verified (failing test → implement → passing test)
4. Do NOT change files outside <your scope>

Return: root cause summary + changes made (file:line for each)

3. Dispatch

With Task tool (Claude Code / Cursor / Codex / OpenCode):

Task("Fix auth module failures in tests/auth/")
Task("Fix pipeline failures in tests/pipeline/")
Task("Fix handler failures in tests/api/")
# Concurrent execution

Without Task tool (Gemini / claude.ai):

# Sequential fallback — run each as a separate focused session
# Session 1: scope=tests/auth/ → systematic-debugging → fix → report
# Session 2: scope=tests/pipeline/ → systematic-debugging → fix → report
# Parallel speed benefit lost. Isolation benefit retained.

4. Integrate and Verify

Goal: confirm fixes don't conflict and the full system is green. Constraints:

• Read each agent's summary before merging
• Verify no two agents touched the same files
• Run full test suite after integration
• For each confirmed bug fixed → trigger pattern-globalize

Common Mistakes

After Integration

For each bug confirmed and fixed by agents:

• Trigger pattern-globalize — one bug found = grep globally for the class
• Run verification-before-completion before declaring all tasks done

Gotchas — Observed Failure Points

<!-- Populated by knowledge-compound after cycles where this skill underperformed --> <!-- Format: [YYYY-MM-DD] What failed | Root cause | What to do instead -->

Evidence Boundary

Subagent reports are not T1 evidence for the parent agent. Treat them as T3 until the parent verifies raw output, diff, or runtime behavior directly. Use shared/subagent-evidence.md and verification-before-completion claim grammar before marking tasks complete.