@darkhunt-security/endpoint-claude-code@0.9.17
Darkhunt endpoint adapter for Claude Code: hook codec, transcript mapper, plugin manifest.
AI Security Review
scanned 4d ago · by lpm-firewall-aiThe package has no install lifecycle scripts and exposes enrollment as an explicit command. Its session capture and transmission to the enrolled Darkhunt tenant are disclosed, with the destination shown during setup and capture configurable.
Decision evidence
public snapshot- The manifest exposes an explicit enrollment command and declares no lifecycle scripts.
- Setup tells the user which endpoint, tenant, workspace, and application will receive sessions and asks them to confirm the destination.
- The README discloses that capture includes session transcripts, thinking blocks, tool calls and results, and subagent transcripts.
- The README says session content is sent to the enrolled Darkhunt tenant and that capture can be disabled in configuration.
- Enrollment sends the Darkhunt API key as a bearer token to the configured API base URL, defaulting to api.darkhunt.ai.
- Capture runs only when runtime settings enable both the endpoint and capture; then it records session events and starts the forwarder.
Source & flagged code
36 flaggedPackage contains a critical-looking secret pattern.
dist/bin/forwarder.mjsView on unpkg · L19382GitHub personal access token in dist/bin/forwarder.mjs
dist/bin/forwarder.mjsView on unpkg · L19382GitHub OAuth access token in dist/bin/forwarder.mjs
dist/bin/forwarder.mjsView on unpkg · L19383Findings
Affected versions and remediation
This report applies to @darkhunt-security/endpoint-claude-code@0.9.17.
See all versions of @darkhunt-security/endpoint-claude-code for other recorded verdicts.
- Review the evidence and your use of @darkhunt-security/endpoint-claude-code@0.9.17 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.