Wiki
Make AI truly understand large codebases: for multi-repository, multi-microservice...
Install
npx skills add Tencent/teamai-cli --skill wiki -g -a claude-code -y- Repository
- Tencent/teamai-cli5.7k
- Category
- AI and agents
- Updated
- 3d ago
- License
- See repository
- Context cost
- Mediumabout 3,000 tokens
- Runs scripts
- Yes, review before use
What it does
Make AI truly understand large codebases: for multi-repository, multi-microservice projects that have evolved over years, run architecture reverse-engineering + a Graph RAG graph + multi-language AST to compress a huge codebase into a structured knowledge base, where every conclusion traces back to a code line and every relation carries a confidence label. Suited to projects with 10+ repositories or microservices that AI cannot understand globally by reading the code directly. Triggers: architecture analysis, architecture reverse-engineering, codebase knowledge base, code-to-knowledge, architecture wiki, large multi-repo codebase. Loaded on demand by the teamai discovery stub.
wiki: AI cognition engineering for large codebases
Prerequisites: an accessible source directory (multiple repositories supported), Python 3, and an installed teamai CLI. The methodology, sub-agent prompts, templates and scripts ship with the CLI. Run
teamai skill path wikito get their absolute path;{SKILL_DIR}in this document refers to that path; a reference file you open on its own writes that directory asSKILL_DIRin braces. Write the knowledge-base documents in Simplified Chinese, as earlier releases did. When updating an existing knowledge base, keep its file names and headings;validate_kb.pyaccepts both the current English and the earlier Chinese headings. The Phase 0 structural baseline usesteamai import --dir. TeamAI does not ship a separate team-wiki CLI. No extra plugin is required.
The problem: large projects (10+ repositories, dozens of microservices, years of iteration) defeat global understanding by AI. The context window cannot hold all the code, component relations are scattered everywhere, and business rules hide deep in call chains. Letting AI read the code directly is both slow (huge token counts) and inaccurate (no global view).
The solution: use architecture reverse-engineering to systematically compress a huge codebase into a structured, verifiable, AI-Native deep knowledge base. Every conclusion traces back to a code line, every relation carries a confidence label, and every update is incrementally verified. AI reads the knowledge base instead of the source, and gains global architecture awareness for about 1/50 of the tokens.
For recalled source freshness, current means only that the cited file matches the scan baseline; check current source when available before relying on claims marked stale, missing or unknown.
Usage
The user states the mode in natural language, or simply says "build a codebase knowledge base":
