claude-code-infinite@1.0.27
Wrapper for Claude Code CLI
AI Security Review
scanned 16d ago · by lpm-firewall-aiThe default MCP integration gives Claude read access to the user's other session history without per-call approval, leaving a concrete agent capability risk that warrants a warning. The package has no npm install hook beyond a prepublishOnly build step.
Decision evidence
public snapshot- A normal ccc launch enables and always loads a first-party MemTree MCP server, and passes its tools as allowed tools; callers can disable it with CCC_MEMTREE_MCP=0, supply their own server, or disallow tools.
- The MCP tools can list the user's other sessions and retrieve exact transcript lines; the handlers route calls to search, read, and list operations.
- The flagged memtree-tools.js locations format search and tail results using local arrays and result fields; they do not establish a separate attack.
- package.json declares prepublishOnly as npm run build, with no install lifecycle hook shown.
- The finder source describes session listing and search through owner-only endpoints using the user's key, which does not reach the MCP process.
- The MCP integration is activated when the user runs ccc and has explicit environment and command-line controls.
- The inspected source does not show an npm install-time attack or a transfer of session data to an unrelated recipient.
Source & flagged code
6 flaggedPackage source references child process execution.
dist/resume-repair-file.jsView on unpkg · L4Findings
Affected versions and remediation
This report applies to claude-code-infinite@1.0.27.
See all versions of claude-code-infinite for other recorded verdicts.
- Review the evidence and your use of claude-code-infinite@1.0.27 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.