@inkeep/open-knowledge@0.76.0
Local-first, agent-friendly Markdown knowledge base with a cross-platform CLI and web editor
AI Security Review
scanned 1mo ago · by lpm-firewall-aiUser-run ok init writes this package into many third-party agent MCP and skill configs; that is explicit dual-use setup, not an npm install hook. Native .node addons and incomplete parse coverage of huge minified files keep residual capability unresolved, so this cannot be marked clean.
Decision evidence
public snapshot- ok init defaults to registering this package as an MCP server in selected editors.
- Init maps Claude, Cursor, and other agents to mcpServers plus skill files such as .claude/skills/open-knowledge/SKILL.md.
- Native NAPI-RS .node addons are loaded to rewrite TOML MCP configs and cannot be fully inspected as source.
- Skill install reporting sends event, source, skill, and agent names to add-skill.vercel.sh.
- The published bins are user-run CLIs; large minified chunks still failed static parse, so leftover eval, dynamic require, and network primitives were not fully traced.
- The bundled eval hit is a sandbox deny list that throws on eval rather than executing it.
- package.json has no preinstall, install, or postinstall scripts.
- GitHub tokens are used for GitHub or GitHub Enterprise device-flow and gh CLI login, not a hidden third-party dump.
- High-similarity files are vendor marked, mermaid, and NAPI-RS loader code.
- Auto-init on other CLI paths runs with mcp disabled.
Source & flagged code
18 flagged · loading sourcePackage source references child process execution.
dist/dist-B1Yrnssl.mjsView on unpkg · L1Findings
Affected versions and remediation
This report applies to @inkeep/open-knowledge@0.76.0.
See all versions of @inkeep/open-knowledge for other recorded verdicts.
- Review the evidence and your use of @inkeep/open-knowledge@0.76.0 before allowing it. Restrict the permissions described in this report.
- Choose an independently verified alternative or release. This report does not establish that other versions are safe.
Evidence last updated: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.