Skill · job · research-agent
Research agent
Long-horizon research with notes on disk, isolated researcher subagents, and a parent that synthesizes 30-line returns. Use for competitive research, literature review, investigations, DeerFlow/Deep Agents — not for side effects or production mutations.
GET /api/canon/skills/research-agent?format=md
Orchestrator-workers. The parent drowns if you dump 40 hits into its window. Filesystem is the workspace.
When
The path is unknown, sources are many, and the deliverable is a brief.
Do
- 01
Parent synthesizes
Decompose, spawn, read summaries. Never paste raw SERP pages into the parent.
- 02
Researchers write notes.md
Fresh window, reduced tools, 30-line return plus a path. Caps on depth/breadth/time.
- 03
Cite
Sources in the schema. Retrieved web is untrusted data.
Don't
- Give researchers deploy keys.
- Treat the parent transcript as the research corpus.
Hard rules
- Notes on disk. Window is a cache.
- No side effects in a research worker.
Refuse
- Context hoarding — Raw HTML, full CI logs, six PDFs, entire JSON envelopes stuffed 'just in case'. The goal is the first thing forgotten.
- Shared mega-context — A crew sharing one bloated window. Specialists drown in each other's work.
- Subagent inherits the parent — Nested loops get every tool and every secret. Isolation was theater.
- Unbounded loop — No max-turns, token, wall-clock, or money budget. The agent rambles until the bill or the context dies.
Load with this
Load next
- Orchestrator–workersSubtasks are not known up front. The parent must not drown in search hits.
- Spawn subagentsProtect the parent window, parallelize, or apply a different skill set.
- Document / RAG agentUsers ask questions over private documents and you must not hallucinate a missing page.
- Own the context windowHour two of a task, or any time the transcript is treated as the product.
Trigger tests
Should fire
- “Build a deep research agent”
- “Competitive landscape with sources”
Should not
- “Refund the customer”
- “Open a PR”