Researcher Skill
Sharded deep research for large repositories using the Qianji runtime (xiuxian-qianji) and a suspend/resume approval loop.
Architecture
┌─────────────┐ ┌────────────────┐ ┌──────────────────┐
│ Setup │ --> │ Architect Plan │ --> │ Await Approval │
│ clone + map │ │ shard proposal │ │ suspend/resume │
└─────────────┘ └────────────────┘ └──────────────────┘
│
▼
┌──────────────────┐
│ Deep Analysis │
│ approved shards │
└──────────────────┘
Command
git_repo_analyer
Core command to execute repository research via Qianji.
Parameters:
repo_url(string, required): Git repository URL to analyze.request(string, optional): Research goal. Default:"Analyze the architecture".action(string, optional):"start"or"approve". Default:"start".session_id(string, required forapprove): Session returned bystart.approved_shards(string, required forapprove): Approved plan JSON string.
Execution model:
action="start":- clones and maps repository,
- asks architect to propose shard plan,
- returns
session_id,proposed_plan, and approval prompt.
action="approve":- resumes same session with approved shard JSON,
- runs deep analysis for approved shards,
- returns final analysis payload.
Output
The command returns structured JSON. Typical fields:
successsession_idmessage/proposed_plan(start phase)analysis_result/full_context(approve phase)
Implementation Notes
- Runtime backend is
xiuxian-qianji(Rust). - Python entrypoint is
scripts/research_entry.py. - Utility functions for clone/map/compress/save are in
scripts/research.py. - Workflow definition is
workflows/repo_analyzer.toml.
Files
researcher/
├── SKILL.md
├── README.md
├── scripts/
│ ├── research.py
│ └── research_entry.py
├── workflows/
│ └── repo_analyzer.toml
└── tests/