Deep research
Cursor agent workflow pack: Spec state machine, ContextPack, and deterministic gates for deep research.
npx -y skills add JunfengRan/dev-env --skill deep-researchAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 13 days oldThe repository was created 13 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
按 spec/research-workflow.yaml 编排多角度深度研究：状态机、ContextPack、subagent barrier、gate 放行。 触发：深入分析、技术选型、竞品对比、源码验证、决策文档、deep research。
SKILL.md
3.0 KB, as published. Nobody here has run it
Deep Research（薄编排层）
阶段定义不在本 SKILL — 唯一真相源是 spec/research-workflow.yaml。
启发式补充（非 Spec 阶段）
研究问题或选型边界仍模糊时，可先用 brainstorm 做几轮澄清与方案对比，再把结论写进 research-brief.json。
- brainstorm 不是 Spec state，不经 gate / reducer
- 用户已明确读者、scope、evidenceTargets 时可跳过
- 交接模板见
brainstorm/references/handoff-templates.md
启动 / 恢复 Run
- 用 CLI 新建 run（推荐）或恢复已有目录（见
spec/run-directory.md）：
node scripts/research-cli.mjs init <run-id> --slug <slug>
# 或: npm run research -- init <run-id> --slug <slug>
- 确认
run-meta.json、state.json、context-pack.json、observations.jsonl、replay-chain.json已生成 - Read Spec + 当前
state.json的currentState（也可用node scripts/research-cli.mjs status）
每轮循环
读 Spec 当前 state → 按 contextSlice 组装 ContextPack 注入
→ 执行（agent / skill / parallel_subagents）
→ 写 artifacts → research-cli advance（gate + reducer + bump）
→ 写 observation（subagent 用 after-subagent-complete）
状态类型
| kind | 行为 |
|---|---|
agent | 主 deep-research agent 执行 |
parallel_subagents | dispatch codebase-verifier；barrier 同步 |
skill | 调用 interpret-tech-notes 或 write-idea-docs |
terminal | 结束 |
禁止
- 未跑 gate / 未
advance就宣称 phase 完成 - 跳过 Spec 未定义的 transition
- 子 agent 结论只写 chat 不写
observations.jsonl - 手改
state.jsonbarrier（应走 reducer / CLI / after-subagent-complete）
References
- read-spec-guide.md
- context-pack-guide.md
- subagent-dispatch-guide.md
- replay-guide.md
- plan-critique-playbook.md
- depth-checklist-decision.md
工具命令
node scripts/research-cli.mjs init <run-id> --slug <slug>
node scripts/research-cli.mjs status [run-dir]
node scripts/research-cli.mjs advance [run-dir]
node scripts/research-cli.mjs apply <run-dir> '<event-json>'
node scripts/gates/<gate>.mjs .research/<run-id>
node plugins/deep-research-gates/scripts/after-subagent-complete.mjs .research/<run-id> <subagentId> <artifact>
node plugins/deep-research-gates/scripts/bump-context-pack.mjs .research/<run-id> <phase>
npm run validate:research-spec