This release includes 6 security fixes for security teams reviewing exposed deployments.
Topics
+12 more
Affected surfaces
Summary
AI summaryProvenance tracking records note origins and flags stale entries, improving trust-aware search.
Full changelog
Memory Integrity
SAME now tracks where your notes came from and whether they're still accurate.
Every saved note records its provenance — the files and notes it was derived from, with SHA256 hashes at capture time. When source files change, SAME detects the divergence and flags affected notes as stale. Stale notes rank lower in search automatically, so your AI gets trustworthy context first.
Highlights
- Provenance tracking —
note_sourcestable records what each note was derived from - Trust-aware retrieval — stale notes rank 25% lower, contradicted notes 60% lower in search results.
trust_statefield on all search results lets agents caveat answers. - Graph 1-hop expansion — top search results are expanded through knowledge graph edges to surface related decisions, references, and dependencies
same kaizen— log friction, bugs, and improvement ideas as you work. Provenance tracking detects when source files change, hinting that items may be addressed.same health— vault health score with trust analysis: validated/stale/unknown counts, stale source detection, actionable recommendations- Crash resilience — PreCompact hook saves checkpoint handoffs before context compaction, not just on session stop
- 5 new MCP tools —
save_kaizen,mem_consolidate,mem_brief,mem_health,mem_forget(17 total)
Also new
same tips— vault hygiene, security, and model selection guidancesame graph enable/disable— toggle graph mode without editing configsame consolidate/same brief— LLM-powered knowledge management [experimental]- Container environment detection (Docker, Kubernetes, Codespaces, Gitpod)
- Graceful Ctrl+C cancellation during reindex and init
- Thinking model compatibility (DeepSeek-R1, QwQ, etc.)
- Windows ARM64 binary
- Human-readable error messages for embedding failures
Fixed
- Windows self-update stale
.oldbackup file handling - Migration failure upgrading from v0.9.1 to v0.10.0
- Graph stats config reading
- URL redirect vulnerability in web dashboard
- MCP SDK bumped to v1.4.0 (security fix)
Security
- Path traversal validation in provenance source recording
- Prompt injection protection for staleness/divergence context tags
- YAML frontmatter injection prevention in kaizen notes
- Path validation in source divergence checks
Full changelog: https://github.com/sgx-labs/statelessagent/blob/main/CHANGELOG.md
Full Changelog: https://github.com/sgx-labs/statelessagent/compare/v0.10.0...v0.12.0
Full Changelog: https://github.com/sgx-labs/statelessagent/compare/v0.10.0...v0.12.0
Security Fixes
- Fixed URL redirect vulnerability in web dashboard
- MCP SDK bumped to v1.4.0 (includes security fix)
- Path traversal validation added for provenance source recording
- Prompt injection protection for staleness/divergence context tags
- YAML frontmatter injection prevention in kaizen notes
- Path validation added to source divergence checks
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About sgx-labs/statelessagent
Memory with provenance tracking — records where agent knowledge originated and detects when sources change. 17 MCP tools for session handoffs, decisions, semantic search, and knowledge graph. Works across Claude Code, Cursor, Windsurf, Codex CLI, and Gemini CLI. Single Go binary, SQLite + vector search, fully local.
Related context
Beta — feedback welcome: [email protected]