Dossier Collect

功能

• Recursive parallel expansion across multiple data sources
• Graph-structured dossier generation with nodes and edges
• Support for various seed types (username, file, symbol, ADR, URL, concept)
• Configurable depth, breadth, sources, and budget
• Output in Markdown, JSON, and persistence to memory store

使用场景

• Investigating a user's online presence or contributions.
• Mapping dependencies and relationships within a codebase or project.
• Discovering all connected entities related to a specific architecture decision record.
• Building a comprehensive profile of a concept by exploring its mentions and related terms.

非目标

• Answering specific questions or performing targeted research (use `deep-research` instead).
• Creating multi-step plans or executing sequential tasks (use `goal-plan` instead).
• Performing deep dives into a single entity without broader exploration.

工作流

1. Detect seed type
2. Pick sources
3. Start trajectory
4. Round 0 fan-out with all source queries
5. Extract entities
6. De-duplicate entities
7. Recurse for new entities until depth/budget exhausted
8. Aggregate nodes and edges into a graph
9. Render artifacts (.md, .json)
10. Persist the dossier
11. End trajectory

实践

• Information gathering
• Graph construction
• Recursive exploration
• Data provenance

先决条件

• Access to `mcp__claude-flow__memory_store`
• Access to `mcp__claude-flow__memory_search`
• Access to `mcp__claude-flow__agentdb_pattern-search`
• Access to `mcp__claude-flow__hooks_intelligence_trajectory-start`
• Access to `WebSearch`, `WebFetch`, `Read`, `Write`, `Grep`, `Glob` tools