Starstruk Logo MarkStarstruk

v12.1.0

thedotmack/claude-memv12.1.0Apr 9, 2026by thedotmack

AI Summary

This release introduces a major Knowledge Agent system that allows users to build queryable AI 'brains' from their Claude observation history, enabling conversational Q&A against compiled corpora with full lifecycle management (build, prime, query, reprime, rebuild, delete). The update includes 6 new MCP tools, 8 HTTP API endpoints, significant security hardening against path traversal and instruction injection, and comprehensive documentation with a Quick Start guide.

Key Highlights

  • New Knowledge Agent system for building queryable AI 'brains' from observation history
  • 6 new MCP tools: build_corpus, list_corpora, prime_corpus, query_corpus, rebuild_corpus, reprime_corpus
  • 8 new HTTP API endpoints on the worker service (/api/corpus/*)
  • Security hardening: path traversal prevention, system prompt injection protection, runtime validation
  • Comprehensive e2e test suite with 31 tests covering full corpus lifecycle

New Features

  • Knowledge Agent system with full lifecycle management (build, prime, query, reprime, rebuild, delete)
  • CorpusBuilder - searches observations, hydrates records, calculates stats, persists to ~/.claude-mem/corpora/
  • CorpusRenderer - renders observations into full-detail prompt text for 1M token context
  • KnowledgeAgent - manages Agent SDK sessions with session resume for multi-turn Q&A
  • Auto-reprime - expired sessions automatically reprimed and retried for session errors
  • Knowledge agent skill (/knowledge) for guided corpus creation
  • Path traversal prevention with alphanumeric name validation + resolved path check
  • System prompt hardened against instruction injection from untrusted corpus content
  • E2e curl wrappers with connect-timeout and transport failure fallback

Full Release Notes

## Knowledge Agents

Build queryable AI "brains" from your claude-mem observation history. Compile a filtered slice of your past work into a corpus, prime it into a Claude session, and ask questions conversationally — getting synthesized, grounded answers instead of raw search results.

### New Features

- **Knowledge Agent system** — full lifecycle: build, prime, query, reprime, rebuild, delete
- **6 new MCP tools**: `build_corpus`, `list_corpora`, `prime_corpus`, `query_corpus`, `rebuild_corpus`, `reprime_corpus`
- **8 new HTTP API endpoints** on the worker service (`/api/corpus/*`)
- **CorpusBuilder** — searches observations, hydrates full records, calculates stats, persists to `~/.claude-mem/corpora/`
- **CorpusRenderer** — renders observations into full-detail prompt text for the 1M token context window
- **KnowledgeAgent** — manages Agent SDK sessions with session resume for multi-turn Q&A
- **Auto-reprime** — expired sessions are automatically reprimed and retried (only for session errors, not all failures)
- **Knowledge agent skill** (`/knowledge-agent`) for guided corpus creation

### Security & Robustness

- Path traversal prevention in CorpusStore (alphanumeric name validation + resolved path check)
- System prompt hardened against instruction injection from untrusted corpus content
- Runtime name validation on all MCP corpus tool handlers
- Question field validated as non-empty string
- Session state only persisted after successful prime (not null on failure)
- Refreshed session_id persisted after query execution
- E2e curl wrappers hardened with connect-timeout and transport failure fallback

### Documentation

- New docs page: Knowledge Agents usage guide with Quick Start, architecture diagram, filter reference, and API reference
- Knowledge agent skill page with workflow examples
- Added to docs navigation

### Testing

- Comprehensive e2e test suite (31 tests) covering full corpus lifecycle

**Full Changelog**: https://github.com/thedotmack/claude-mem/compare/v12.0.1...v12.1.0