AgentSkillsCN

scout

在代码库中搜索相关文件与上下文信息；当被要求定位文件或明确工作方向时使用。

SKILL.md
--- frontmatter
name: scout
description: Search the codebase for relevant files and context; use when asked to locate files or understand where to work.

Scout

Overview

Use CLI subagents to find relevant files and optionally gather documentation.

Inputs

  • •Search query
  • •Scale (1-6, default 2)

Scale Map

  • •1: Gemini Flash
  • •2: + Gemini Lite
  • •3: + Codex
  • •4: + Claude Haiku
  • •5: + 2x Gemini Flash doc scouts
  • •6: + 2x Codex doc scouts

Workflow

  1. •Parse query and scale.
  2. •Run the subagent commands for the chosen scale (parallel if possible).
  3. •Collect outputs, dedupe, and rank files by consensus.
  4. •If scale >= 5, save docs under aidocs/ and aidocs/context/.

Subagent Commands

Gemini Flash (codebase)

bash
CLOUDSDK_CORE_PROJECT="" GOOGLE_CLOUD_PROJECT="" GCLOUD_PROJECT="" GEMINI_API_KEY=${GEMINI_API_KEY} \
  gemini -m gemini-3-flash-preview -o text "Search the codebase for files related to: {QUERY}. Return file paths with line ranges."

Gemini Lite (codebase)

bash
CLOUDSDK_CORE_PROJECT="" GOOGLE_CLOUD_PROJECT="" GCLOUD_PROJECT="" GEMINI_API_KEY=${GEMINI_API_KEY} \
  gemini -m gemini-2.5-flash-lite -o text "Quickly identify files for: {QUERY}. List file paths with line ranges."

Codex (codebase)

bash
codex exec -m gpt-5.2 -s read-only -c model_reasoning_effort="low" --skip-git-repo-check \
  "Locate files for: {QUERY}. Output file paths with line numbers."

Claude Haiku (codebase)

bash
claude --model haiku -p "Find files for: {QUERY}. Return file paths with line ranges only."

Gemini Flash (docs 1)

bash
CLOUDSDK_CORE_PROJECT="" GOOGLE_CLOUD_PROJECT="" GCLOUD_PROJECT="" GEMINI_API_KEY=${GEMINI_API_KEY} \
  gemini -m gemini-3-flash-preview -o text "Search the internet for official documentation related to: {QUERY}. Return URLs and short summaries."

Gemini Flash (docs 2)

bash
CLOUDSDK_CORE_PROJECT="" GOOGLE_CLOUD_PROJECT="" GCLOUD_PROJECT="" GEMINI_API_KEY=${GEMINI_API_KEY} \
  gemini -m gemini-3-flash-preview -o text "Find tutorials and best practices for: {QUERY}. Return URLs and short summaries."

Codex (docs 1)

bash
codex exec -m gpt-5.2 -c model_reasoning_effort="low" --skip-git-repo-check \
  "Search online docs for: {QUERY}. Return URLs and summaries."

Codex (docs 2)

bash
codex exec -m gpt-5.2 -c model_reasoning_effort="low" --skip-git-repo-check \
  "Find code examples and guides for: {QUERY}. Return URLs and summaries."

Docs Saving (scale >= 5)

bash
mkdir aidocs
mkdir aidocs\context

Save results to aidocs/<query-slug>-docs.md with URLs and key findings.

Output

  • •Ranked file list with brief notes
  • •Optional docs summary file path