30 lines
1.4 KiB
Markdown
30 lines
1.4 KiB
Markdown
---
|
|
name: local-scout
|
|
description: "Delegates codebase exploration to the local scout subagent (runs on a fast local Qwen model with QMD + opty tools). Load this skill only when the user explicitly asks to use scout, use local scout, or scout a task. Do NOT load automatically for general exploration — only when scout is explicitly requested."
|
|
---
|
|
|
|
# Local Scout
|
|
|
|
Delegate codebase exploration to the `scout` subagent, which runs on a fast local model (Qwen) augmented with semantic search via QMD and HDC-indexed context retrieval via opty. It is cheap, fast, and keeps the main context clean.
|
|
|
|
## When to use
|
|
|
|
- User says "use scout to find …", "scout: …", or "use local scout"
|
|
- You need to gather broad codebase context before planning
|
|
- The task is primarily "look around the codebase" rather than making precise edits
|
|
|
|
## How to invoke
|
|
|
|
```javascript
|
|
subagent({ agent: "scout", task: "Find and summarize the authentication flow" })
|
|
```
|
|
|
|
The scout writes its findings to `context.md` and returns a summary. Use the summary or read `context.md` for the full structured output.
|
|
|
|
## Tips
|
|
|
|
- Be specific in the task description — the scout infers thoroughness from it
|
|
- For deep traces, prefix with "Thorough:" e.g. `"Thorough: trace all usages of X"`
|
|
- For quick lookups, prefix with "Quick:" e.g. `"Quick: where is the config loaded?"`
|
|
- Do your own reading only when you need precise line-level content to reference in your response
|