The root README is the landing page. This index points to the canonical technical references; update the owning page rather than duplicating volatile behavior elsewhere.
- CLI reference — command/path contracts, installation targets, daemon/watch lifecycle, configuration, and diagnostics
- MCP reference — tool schemas, visible/known tools, project resolution, stdio and streamable HTTP, protocol behavior, and client configuration
- Browser viewer — the local read-only viewer (preview): usage, boundary, live channel, API, and differences from upstream
- Supported languages — user-facing language coverage, extensions, extraction tiers, and static-analysis boundaries
- Godot static analysis — scenes, resources, scripts, autoloads, honesty signals, and known runtime boundaries
- Troubleshooting — opt-in JSONL diagnostics and evidence to include in a report
- Architecture — workspace dependency graph and the extraction, storage, resolution, query, MCP, daemon, and watcher flow
- Data model — SQLite/FTS5 tables, indexes, migrations, and connection policy
- Equivalence oracle — stable IDs, canonicalization tiers, golden corpora, and regeneration procedures
- Grammar ABI manifest — grammar/custom extractor ownership and contributor checks
- Embedded extraction — Vue, Svelte, Astro, Razor, Liquid, MyBatis XML, and DFM/FMX delegation
- Benchmark methodology and current result status — how performance evidence is collected and what is currently safe to claim
- Contributing
- Security policy
- Code of Conduct
- Agent contributor contract
- Upstream synchronization ledger
- Dated upstream audits
English is canonical for deep technical references. The Chinese README mirrors the landing page, commands, security posture, and navigation; it does not duplicate every implementation reference.