메뉴
Parallel context-gathering for a code area. Use when you need to understand a module, feature, or subsystem and would otherwise read 3+ files sequentially — dispatches two agents in parallel to map structure and test coverage in one wave.
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
Orchestrates safe, large-scale structural changes across a codebase — symbol renames, API migrations, pattern standardizations, layer restructurings. Enumerates all affected sites, groups them into dependency layers via the DAG executor, applies changes in parallel per layer with worktree isolation, and verifies behavioral preservation at each layer boundary before proceeding. Exits immediately on first regression, surfacing the exact site, diff, and behavioral delta before any commit.
Parallel root-cause analysis for bugs and failing tests. Use when a test fails, a bug is reported, or behavior is unexplained — dispatches sub-agents to form and validate hypotheses in isolated worktrees.
Set up a scheduled headless afk run that pushes a summary to Telegram. Use when the user wants to automate a recurring task via the afk daemon scheduler (cron) with push-notified results.
Dispatches two sub-agents in parallel to gather external and local context for the current task.
Dispatch a parallel adversarial verifier wave after any high-stakes sub-agent investigation (code reviews, audits, findings reports, large refactors) — or whenever a sub-agent asserts a claim with high-confidence language (confident, certain, clearly, ≥80%), since confidence is a trigger, not a verdict. Shadow verifiers independently re-derive 2–3 key claims from scratch using tool calls only, returning CONFIRMED/REFUTED/UNVERIFIABLE, and flag disagreements before the user acts. Use when sub-agent output will drive decisions, file changes, commits, or external side-effects.
Discovers incidental complexity, duplication, and dead code in a codebase and produces a ranked, behavior-preserving reduction plan — optionally applying safe changes. Dispatches four parallel read-only discovery lenses (clone detection, dead code, complexity hotspots, wrong abstraction), synthesizes into a prioritized reduction plan, and gates apply mode behind /refactor and a hard test check.
| name | gather |
| description | Parallel context-gathering for a code area. Use when you need to understand a module, feature, or subsystem and would otherwise read 3+ files sequentially — dispatches two agents in parallel to map structure and test coverage in one wave. |
| context | load |
You MUST emit exactly two agent tool_use blocks in a single response turn — both calls in the same assistant message, before either result arrives. Do not dispatch the second agent in a later turn after seeing the first agent's reply. Do not dispatch three agents. Do not dispatch one.
Correct shape of your next response:
<assistant turn>
<tool_use name="agent" id="…"> Structure Agent prompt </tool_use>
<tool_use name="agent" id="…"> Test Agent prompt </tool_use>
</assistant turn>
If you find yourself about to send a single agent call and wait, stop — that is the failure mode this skill exists to prevent.
When understanding a task requires reading multiple related files (imports, callers, tests, configs, types), dispatch these two — concurrently, per the protocol above:
Structure Agent (Explore, thoroughness matched to scope) — Find and read the target file(s), all direct imports, callers, and config references. Return:
files_read: absolute paths examinedcall_graph: how components connect (one paragraph)public_interfaces: function signatures, types, or contracts that govern the areaentry_points: where control flow entersTest Agent (Explore, "medium") — Find test files that exercise the target area, read them, identify what paths are covered and what's missing. Return:
test_files: absolute paths of relevant testscoverage_summary: what behaviors/branches tests exerciseuntested_paths: code paths with no test coverageWhen both return, merge into a unified context map. If either agent's output has gaps (e.g., Structure Agent missed config, Test Agent found no tests), issue one targeted follow-up Read — do not re-dispatch.