메뉴
Use when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Ingest a raw source into the project's OKF knowledge bundle — distill it into concept pages, refresh the index, and log the change. keel's compounding "LLM wiki" loop.
Use when implementation is complete, gates pass, and you need to decide how to integrate the work — presents structured options (merge, PR, keep, discard) and handles the chosen workflow, including worktree cleanup. Closes the SDD cycle after review-and-simplify.
Use when executing implementation plans with independent tasks in the current session
Orchestrates the implement-feature, evaluator, fix-runner loop for the active feature until all Criterios de Aceitacao are met and quality gates are green.
Use after the plan is approved to break the active feature's plan into a concrete task checklist, written into specs/<feature>/tasks.md from the tasks template.
Use while implementing a feature that has retries, queues, background jobs, external calls, or any critical path an on-call engineer will need to reason about in production. Invoke when adding a new endpoint, dependency, or async flow, and before considering such a feature done — instrumentation is developed alongside the code, like tests, not bolted on after.
| name | dispatching-parallel-agents |
| description | Use when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies |
You delegate tasks to specialized agents with isolated context. By precisely crafting their instructions and context, you ensure they stay focused and succeed at their task. They should never inherit your session's context or history — you construct exactly what they need. This also preserves your own context for coordination work.
When you have multiple unrelated failures (different test files, different subsystems, different bugs), investigating them sequentially wastes time. Each investigation is independent and can happen in parallel.
Core principle: Dispatch one agent per independent problem domain. Let them work concurrently.
digraph when_to_use {
"Multiple failures?" [shape=diamond];
"Are they independent?" [shape=diamond];
"Single agent investigates all" [shape=box];
"One agent per problem domain" [shape=box];
"Can they work in parallel?" [shape=diamond];
"Sequential agents" [shape=box];
"Parallel dispatch" [shape=box];
"Multiple failures?" -> "Are they independent?" [label="yes"];
"Are they independent?" -> "Single agent investigates all" [label="no - related"];
"Are they independent?" -> "Can they work in parallel?" [label="yes"];
"Can they work in parallel?" -> "Parallel dispatch" [label="yes"];
"Can they work in parallel?" -> "Sequential agents" [label="no - shared state"];
}
Use when:
Don't use when:
Group failures by what's broken:
Each domain is independent - fixing tool approval doesn't affect abort tests.
Each agent gets:
Issue all three subagent dispatches in the same response — they run in parallel:
Subagent (general-purpose): "Fix agent-tool-abort.test.ts failures"
Subagent (general-purpose): "Fix batch-completion-behavior.test.ts failures"
Subagent (general-purpose): "Fix tool-approval-race-conditions.test.ts failures"
# All three run concurrently.
Multiple dispatch calls in one response = parallel execution. One per response = sequential.
When agents return:
Good agent prompts are:
Fix the 3 failing tests in src/agents/agent-tool-abort.test.ts:
1. "should abort tool with partial output capture" - expects 'interrupted at' in message
2. "should handle mixed completed and aborted tools" - fast tool aborted instead of completed
3. "should properly track pendingToolCount" - expects 3 results but gets 0
These are timing/race condition issues. Your task:
1. Read the test file and understand what each test verifies
2. Identify root cause - timing issues or actual bugs?
3. Fix by:
- Replacing arbitrary timeouts with event-based waiting
- Fixing bugs in abort implementation if found
- Adjusting test expectations if testing changed behavior
Do NOT just increase timeouts - find the real issue.
Return: Summary of what you found and what you fixed.
❌ Too broad: "Fix all the tests" - agent gets lost ✅ Specific: "Fix agent-tool-abort.test.ts" - focused scope
❌ No context: "Fix the race condition" - agent doesn't know where ✅ Context: Paste the error messages and test names
❌ No constraints: Agent might refactor everything ✅ Constraints: "Do NOT change production code" or "Fix tests only"
❌ Vague output: "Fix it" - you don't know what changed ✅ Specific: "Return summary of root cause and changes"
Related failures: Fixing one might fix others - investigate together first Need full context: Understanding requires seeing entire system Exploratory debugging: You don't know what's broken yet Shared state: Agents would interfere (editing same files, using same resources)
Scenario: 6 test failures across 3 files after major refactoring
Failures:
Decision: Independent domains - abort logic separate from batch completion separate from race conditions
Dispatch:
Agent 1 → Fix agent-tool-abort.test.ts
Agent 2 → Fix batch-completion-behavior.test.ts
Agent 3 → Fix tool-approval-race-conditions.test.ts
Results:
Integration: All fixes independent, no conflicts, full suite green
Time saved: 3 problems solved in parallel vs sequentially
After agents return:
From debugging session (2025-10-03):
Delegate only bulk, context-polluting, or genuinely parallelizable work. A task that is small or highly context-dependent on the current conversation should run inline instead — dispatch overhead (constructing an isolated prompt, waiting for the round trip, re-integrating the result) costs more than just doing the work directly. Reserve dispatch for cases where the work would otherwise flood your own context with file reads, search results, or long tool output, or where independent pieces can genuinely proceed at the same time.
Pick the cheapest model tier that can do the task perfectly without forcing a costly re-run. Mechanical or bulk work — renames, fixed-pattern edits, transcription against a fully-specified brief — routes to the cheapest available tier. Reasoning, synthesis, or judgment-heavy work — architectural tradeoffs, ambiguous debugging, cross-file integration — routes to the top tier. A cheap model that takes three extra turns to flail toward the right answer often costs more in wall-clock and context than dispatching the right tier the first time, so size the model to the task's actual difficulty, not to a default.