| 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 |
deep-research
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.
State
- Input:
question.
- Outputs:
answer, tool_trace (every tool call the agent made),
verdict, score, rationale.
Use
- Standalone:
stargraph run graph.yaml (network tools need the
tools extra installed; set the LM via --lm-model/--lm-url or
per-node config.model).
- As a subgraph:
kind: subgraph, spec: <path to graph.yaml>;
config.max_steps bounds the research loop.
Tuning
Edit graph.yaml: the check rubric (what counts as complete), the
tool allowlist, or max_iters per research pass.
