Cruxhive — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited Cruxhive (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
The hive mind for your team's AI context — approved, versioned, trusted.
Your coding AI forgets everything between sessions. Your teammates' AIs forget everything too. CruxHive fixes that — with a human approval gate, hybrid semantic search, and a knowledge base that compounds over time.
npx cruxhive initOne human-curated knowledge base. Every AI tool reads from it — Claude Code, OpenCode, Cursor, Windsurf, Gemini CLI. Local SQLite, MIT licensed, zero cloud dependency.
# One command — installs the MCP server (via uv) and wires up your project
npx cruxhive initOr install the engine directly:
uv tool install cruxhive-mcp # Python 3.11+Wire into any MCP client — add to .mcp.json:
{
"mcpServers": {
"cruxhive": { "command": "uvx", "args": ["cruxhive-mcp"], "type": "stdio" }
}
}AI tools never write to your knowledge base directly. They propose; a human approves.
context_propose → the entry lands in .llm/pending/.cruxhive review → approve or reject, with full git history.Solo developer? cruxhive solo auto-approves your own proposals.
init Bootstrap CruxHive in the current project
index Index .llm/ markdown files into the local knowledge base
propose Propose a new knowledge entry for human review
review Interactively approve or reject pending proposals
sync Sync org-layer context from the configured remote
health Knowledge base health summary
stats Usage observability — searches, hit rate, gaps, by AI tool
digest Weekly markdown digest — gaps, decayed entries, queue health
status One-line health summary (use --quiet for hooks)
doctor Diagnose setup — symlinks, hooks, slash commands
workspace Cross-project rollup — aggregate KPIs across all projects
direnv Write a .envrc that auto-logs sessions for Cursor/Windsurf/Gemini
solo Enable/disable solo mode — auto-approve your own proposals
ui Open the approval-queue dashboard (localhost:3847)| Tool | What it does |
|---|---|
context_search | Hybrid search (FTS5 BM25 + optional sqlite-vec, RRF fusion, entity boost) |
context_workspace_search | Same, across every configured project |
context_propose | Submit a knowledge entry for human review |
context_review / context_approve / context_reject | The approval gate |
context_index | (Re)index .llm/ markdown into the local store |
context_check_faithfulness | NLI-based contradiction detection on captured context |
context_radar | Scan recent commits, map to plan areas, surface uncovered work |
context_next_slice · context_write_plan · context_sync_memory | Planning + org-layer sync |
Published. cruxhive-mcp on PyPI · @cruxhive/cli on npm. MIT, 0 cloud dependencies.
Engineering work has 7 stages. Every AI tool covers stages 1–6. Stage 7 — Capture — has no tooling. That's where the value leaks out: hard-won context dies at the end of every session.
CruxHive is infrastructure for Stage 7.
| AGENTS.md | Grov | Kiro | CruxHive | |
|---|---|---|---|---|
| Human approval workflow | ✗ | ✗ | ✗ | ✓ |
| Git versioning + audit trail | partial | ✗ | ✗ | ✓ |
| Semantic search (hybrid, local) | ✗ | ✓ (API) | ✗ | ✓ |
| Org-layer (cross-project) | ✗ | ✗ | ✗ | ✓ |
| Faithfulness detection | ✗ | ✗ | ✗ | ✓ |
| Tool-agnostic via MCP | partial | partial | Kiro only | ✓ |
MIT
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.