[ROADMAP] Phase 3 — Complete (Navigator + Guardian all-in-one, drop codebase-memory-mcp)

[Wolfvin]

Goal

After Phase 1 (#20) and Phase 2 (#22), CodeLens becomes the only tool an AI agent needs:
Navigator + Explorer + Guardian + Inspector — all in one.

No more need for codebase-memory-mcp or any external structural analysis tool.

Checklist

• [FEATURE] Cypher-like graph query engine for MCP tool query_graph #9 — Cypher-like query engine (query_graph MCP tool)
• [FEATURE] Semantic vector search via semantic_query MCP tool #11 — Semantic vector search (semantic_query MCP tool)
• [FEATURE] 158-language support via universal tree-sitter grammar loader #18 — 50+ language support via universal grammar loader
• [FEATURE] Cross-repo intelligence: link edges across multiple indexed repos #15 — Cross-repo intelligence (microservice / monorepo)
• [FEATURE] Team-shared graph snapshot: commit compressed index to repo #12 — Team-shared compressed snapshot (commit index to repo)
• [FEATURE] Architecture Decision Records (ADR) via manage_adr MCP tool #16 — Architecture Decision Records (persistent agent memory)

Definition of Done — the 'drop codebase-memory-mcp' milestone

An AI agent using only CodeLens can:

1. Orient in a new codebase in 1 call (get_architecture)
2. Find any symbol by name, pattern, or semantic meaning
3. Trace call chains across files, packages, and repos
4. Know the blast radius of any change before making it
5. Guard every write: no duplicates, no overwrites, no dead code added
6. Detect secrets, CVEs, taint flows, and complexity issues
7. Do all of the above in <10k tokens per session on a 10k-file repo

Competitive position

Capability	codebase-memory-mcp	CodeLens Phase 3
Structural navigation	✅	✅
Cypher-like queries	✅	✅
Semantic search	✅	✅
150+ languages	✅	✅
Guard hooks (pre-write)	❌	✅
Taint analysis	❌	✅
CVE scanning	❌	✅
Dead code detection	basic	✅
ADR management	✅	✅
Single binary	✅	via wheel
Python-native install	❌	✅