AgentSkillsCN

codebase-context

通过每文件的API清单与导入图，生成架构文档与LLM上下文。当用户想要记录代码库结构、创建架构图、为AI助手生成上下文，或了解模块依赖与代码组织时使用。

SKILL.md
--- frontmatter
name: codebase-context
description: >
  Generate architecture documentation and LLM context with per-file API inventory
  and import graph. Use when the user wants to document the codebase structure,
  create architecture diagrams, generate context for AI assistants, or understand
  module dependencies and code organization.

Codebase Context Skill

When to Use

Activate when the user wants to:

  • •Generate or update architecture documentation
  • •Create LLM-friendly context files for the codebase
  • •Understand module dependencies and code organization
  • •Produce API inventories or import graphs

User Preferences

Before generating context, resolve the user's Profile:

  1. •Check for .github/copilot-preferences.yml or .vscode/copilot-preferences.yml in the workspace.
  2. •Apply any custom documentation conventions or output format preferences.

Inputs (optional; defaults in parentheses)

  • •Depth: standard | deep | light (default: standard)
  • •Max files to inventory: 300
  • •Max functions per file: 200

Policy

  • •No CLIs or network. Edit files directly; then print a change summary.
  • •Be concise and DRY (avoid repeating the same info in multiple sections).
  • •If a section would exceed practical size, summarize patterns and provide 5–10 exemplars.

Scope

  • •Focus on source files; skip generated/third-party/build artifacts.
  • •Exclude: .git/**, **/node_modules/**, **/vendor/**, **/build/**, **/dist/**, **/target/**, **/.next/**, **/.nuxt/**, **/.vercel/**, **/coverage/**, **/__pycache__/**, **/.venv/**, **/.mypy_cache/**, **/.idea/**, **/.vscode/**, **/*.min.*, **/*.map, **/*.bundle.*, **/*.{png,jpg,jpeg,gif,ico,svg,woff,woff2,ttf,eot,exe,dll,so,dylib}.

Steps

  1. •

    Overview / Architecture

    • •Identify modules, layers, and boundaries; note entry points.
    • •Produce one ASCII map and one Mermaid flowchart (flowchart TD) for high-level data/control flow.
    • •Include a sequenceDiagram for one critical request path.
  2. •

    Per-directory Snapshot (1–2 lines each)

    • •Purpose, major components, notable risks.
  3. •

    Per-file API Inventory (budgeted)

    • •For each file (up to max), list exported/public functions and important reused helpers.
    • •Signature format: name(param: type, ...): returnType.
    • •For classes: list key methods and constructor params with one-line purposes.
    • •If a file exceeds max functions, summarize patterns and show 5–10 exemplars.
    • •Mark generated/boilerplate files as such and skip details.
  4. •

    Connections (imports/uses)

    • •Build an adjacency list (normalized module paths, no extensions).
    • •Provide a compact ASCII graph and a Mermaid flowchart TD of module dependencies.
    • •Call out high-fan-in hotspots.
  5. •

    Style, Data, and Risks

    • •Infer code style guide from configs and exemplars.
    • •Summarize data formats (schemas, JSON shapes, wire protocols) with file references.
    • •List top 5 risks/TODOs (security/perf/maintainability) and coverage gaps.

Outputs

  • •Write docs/architecture.md with: Overview, ASCII map, Mermaid flow, sequenceDiagram, module/layer narrative.
  • •Write docs/llms-context.md with: Per-directory snapshot, per-file API inventory, connections, style/data, risks/TODOs.
  • •Keep architecture.md high-level; keep details in llms-context.md to avoid duplication.

Guardrails

  • •Do not call CLIs or the network.
  • •Prefer #selection or #file to seed context; expand to @workspace only as needed.