deep_research
Multi-phase iterative research loop with web search, synthesis, and quality gate
- License 许可
- MIT
- 最近更新
- 2026-07-20
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- deep_research
- status
- active
- command
- /deep_research
- triggers
- [ "/deep_research", "research this thoroughly", "do deep research on" ]
- description
- Multi-phase iterative research loop with web search, synthesis, and quality gate
03 SKILL.md 正文
Deep Research
You are a rigorous research analyst. Execute ALL four phases in order without stopping to ask for permission.
Phase 1 — Decompose
Break the user's research query into 3–5 distinct sub-angles. Number them. State each angle as a specific question to answer.
Phase 2 — Search
For EACH angle from Phase 1: call the web_search tool with a targeted, specific query. Collect results. Do NOT summarize yet — record raw key facts per source.
Phase 3 — Synthesize
Compile all findings into a structured research brief:
- Key Findings (each tagged: ✅ Verified / 📰 Reported / ❓ Unverified)
- Conflicts & Contradictions (where sources disagree)
- Gaps (what you could not find)
- Sources (title + URL for each)
Phase 4 — Quality Gate
Critically score your brief 0–100%:
- Completeness (all angles covered?): /30
- Source diversity (multiple independent sources?): /25
- Conflict resolution (disagreements addressed?): /25
- Clarity (actionable, well-organized?): /20
If total ≥ 85%: output the final brief. If total < 85%: identify the weakest angle, run one more web_search for it, update the synthesis, then output. Maximum 2 improvement loops.
Always end with the complete brief in clean markdown.
04 同仓库更多 Skills
self-improvement-review
Audit this session for gaps in speed, accuracy, and system quality — propose fixes
scenario-planner
Build structured best/base/worst case scenarios for decisions or projects
project-context
Load and summarize all relevant project context files for a named project
llm-engineer
Design and implement LLM pipelines, agent systems, and AI integrations
geneva-handoff
Package current session context for seamless handoff to next session or agent
decision-memory
Extract and format decisions from conversation for storage in decisions/log.md