メニュー
Use when the operator wants parallel sub-agent research aggregated into a structured report. Spawns up to 4 independent workers each investigating a different angle, then synthesizes findings for operator review. Research feeds decisions, does not trigger execution.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Use when a long-form manuscript (book chapter, ebook, multi-chapter playbook, long-form digital product) needs an editing pass to identify and remove AI writing tells. Sweeps across 8 pattern categories, assembles a structured edit packet for operator review, and applies approved edits.
Use when you have spare capacity and want to set your system improving without active attention. Operator-invoked only, never scheduled. Safe, additive, reversible hygiene work runs under a threshold model with a permanent floor of actions the agent never takes on its own.
Use when archiving several workspace artifacts at once with per-artifact approval gates before any archive move executes.
Use when deciding whether a piece of deterministic, no-judgment work should route to a secondary AI provider instead of your primary interface. Covers the provider catalog, the five-condition eligibility test and the dispatch script.
Retire a stalled or indefinitely-deferred plan without losing its value. Candidate scan, harvest, distribute, sweep, archive and codify.
Automated pull request review for your repos. Five parallel agents, confidence scoring, convention-file compliance, and GitHub comment posting.
| name | researcher |
| description | Use when the operator wants parallel sub-agent research aggregated into a structured report. Spawns up to 4 independent workers each investigating a different angle, then synthesizes findings for operator review. Research feeds decisions, does not trigger execution. |
| status | active |
| version | 1.2 |
Purpose: Parallel sub-agent research on a topic, aggregated into a structured report. Spawns up to 4 independent workers, each investigating a different angle, then synthesizes findings into a single report for operator review. Research feeds decision-making. It does not trigger execution.
Triggered by: "research...", "give me a deep dive on...", "before we plan this, let me understand...", "what's the landscape for..."
Phase 1: Clarify angles
If the research request does not specify angles, propose them before spawning:
"I'll research [topic] from these angles: [A], [B], [C], [D]. Confirm or adjust before I begin."
Typical angle patterns:
Phase 2: Spawn parallel workers
Spawn up to 4 sub-agents simultaneously. Each worker gets:
Workers are independent. They do not communicate with each other. Only their final outputs return.
Phase 3: Synthesize
Aggregate all worker outputs into a single structured report:
## Research Report: [Topic]
Date: [YYYY-MM-DD]
### Summary
[2-4 sentence synthesis across all angles]
### Findings by Angle
#### [Angle A]
...
#### [Angle B]
...
### Recommended Actions / Decision Points
[Proposed next steps for operator review. Never auto-implement]
### Sources
[Files read, repos checked, protocols referenced]
Phase 4: Present
Present the report. Do not act on findings. The operator decides what moves forward.
Parallel research workers do not all have to route to the same model or tool. The principle: match each angle's question shape to the substrate best suited for it.
| Angle type | Default substrate | Why |
|---|---|---|
| Synthesis-heavy (compare sources, identify patterns) | Main reasoning model | Reasoning across sources |
| Web-current (post-training-cutoff facts, live state) | Real-time web search tool | Live data beyond model cutoff |
| Mechanical multi-source pull (changelogs, API diffs, structured extraction) | Smaller/cheaper model | No-judgment volume work |
| Adversarial check of a claim | Challenger model | Independent cross-model pressure |
| Voice or intent judgment | Orchestrator model | Register and governance coherence |
Synthesis pass: after all workers return, the orchestrator synthesizes across substrates. Findings from different model types carry different confidence textures; name those differences in the synthesis.
Audit gate: when the synthesized report will inform a major plan or decision, offer a challenger-model review of the synthesis before closing. One line: "Challenger pass warranted?" Do not require it.
This skill spawns parallel sub-agents for research. Sub-agents are intelligence-gathering only. They do not implement, write to canonical files or trigger external actions. Every report returns to the operator for review and decision. That boundary is non-negotiable.
Dispatch the cheapest model that does the job well. Before each delegated step, ask whether a smaller model would produce equivalent output.
| Work type | Model |
|---|---|
| Mechanical lookups, deterministic commands, structured extraction against a spec | Haiku |
| Multi-step synthesis, drafting, diagnosis, most worker dispatch | Sonnet |
| Architectural judgment, plan design, judgment-dense synthesis | Opus |
Per-phase defaults: Phase 1 (clarify angles, operator-facing) and Phase 4 (present, operator-facing) route to Opus. Phase 2 (spawn workers) routes per the Cross-Model Perspectives substrate table above; use Sonnet as the fallback for any angle that does not fit a named type. Phase 3 (synthesize across substrates) routes to Opus because cross-model synthesis carries the highest judgment density.
Set the model explicitly on every subagent dispatch. Never silently inherit the top tier.
Source Harvest is the gateway skill for systematic pattern extraction from external repos and tools. Researcher gathers findings; Source Harvest extracts patterns once you've decided which sources to dig into. Many users adopt Source Harvest first, then layer additional skills like this one on top.
If Source Harvest isn't installed yet: Install Source Harvest via IGOS.