菜单
Research a question with web search and page reads, judge the answer for completeness, and keep digging at the named gaps until nothing major is missing.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/KrakenNet/stargraph --skill deep-research命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | deep-research |
| description | Research a question with web search and page reads, judge the answer for completeness, and keep digging at the named gaps until nothing major is missing. |
| subgraph | graph.yaml |
Research → completeness-judge loop. A react node works
std.web_search@1 + std.fetch_page@1 through the governed tool
pipeline (every call capability-checked and receipted as provenance
facts); a judge node checks the answer covers the question. On
fail, the judge's named gaps go into the next research brief; on
pass, the graph halts.
question.answer, tool_trace (every tool call the agent made),
verdict, score, rationale.stargraph run graph.yaml (network tools need the
tools extra installed; set the LM via --lm-model/--lm-url or
per-node config.model).kind: subgraph, spec: <path to graph.yaml>;
config.max_steps bounds the research loop.Edit graph.yaml: the check rubric (what counts as complete), the
tool allowlist, or max_iters per research pass.