AgentSkillsCN

researching-codebase

在规划之前，先对代码库进行全面调研。在开展任何非 trivial 的实现任务之前，先单独收集并梳理相关背景信息。

SKILL.md
--- frontmatter
name: researching-codebase
description: Investigates codebase before planning. Use before any non-trivial implementation task to gather context in isolation.
argument-hint: [topic-or-question]
context: fork
agent: Explore
allowed-tools: Read, Grep, Glob

Codebase Research (ACE-FCA)

Query: $ARGUMENTS

Gathers codebase context in isolation before planning. Prevents search artifacts from polluting main context.

Core Principles

  1. •Documentation-only - Describe what exists, where, and how it works
  2. •No evaluations - Never suggest improvements or critique implementation
  3. •Evidence-based - Provide file paths, line numbers, and code references
  4. •Isolation - Research runs in fork context; return only distilled findings

When to Use

  • •Before planning non-trivial implementations
  • •When unfamiliar with relevant codebase areas
  • •Before architectural decisions

Workflow

  1. •Read mentioned files first - If specific files mentioned, read completely before exploring
  2. •Decompose the question - Break query into researchable components
  3. •Explore codebase - Investigate architecture, patterns, constraints
  4. •Identify scope - Determine relevant areas based on findings
  5. •Distill - Return structured summary using output format below

Output Format

Follow ACE-FCA quality equation: Correct + Complete + Minimal noise

markdown
---
research_query: "<original question>"
timestamp: "<ISO 8601>"
files_examined: <count>
---

## Key Files

| File | Purpose | Key Lines |
|------|---------|-----------|
| `path/to/file.ext` | Brief purpose | L42-58 |

## Patterns

- **Pattern name**: Description with file reference (`path:line`)

## Constraints

- Constraint with evidence (`path:line`)

Evidence Requirements

Every claim must include:

  • •File path: Exact location (src/auth/login.ts)
  • •Line numbers: Specific lines (L42-58 or L127)
  • •Code reference: Function/class name when relevant

Bad: "Authentication uses JWT tokens" Good: "Authentication uses JWT tokens (src/auth/jwt.ts:L23-45, verifyToken function)"

References

See .claude/rules/context-management.md and .claude/rules/core-principles.md.