claude-devtools-inspector

Fail

Audited by Gen Agent Trust Hub on May 17, 2026

Risk Level: HIGHCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The skill documentation details how the tool accesses highly sensitive files, including the user's SSH configuration (~/.ssh/config) and private keys (~/.ssh/id_rsa), to monitor remote sessions. Access to private keys is a critical security risk.
  • [EXTERNAL_DOWNLOADS]: Users are instructed to install software from untrusted sources, including a non-verified GitHub repository (github.com/matt1398/claude-devtools) and an unverified Homebrew cask.
  • [COMMAND_EXECUTION]: The documentation provides commands for the user to execute that modify filesystem permissions (chmod 755 ~/.claude) and bypass macOS security features (xattr -d com.apple.quarantine) for unverified applications.
  • [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting untrusted session logs from ~/.claude/sessions/ and ~/.claude/projects/. Malicious content within these logs could influence the tool's behavior or the agent's analysis.
  • Ingestion points: Reads session log files (JSONL) and project memory markdown files from the local filesystem.
  • Boundary markers: There are no instructions or delimiters provided to ensure the agent ignores or sanitizes embedded instructions within the logs.
  • Capability inventory: The tool performs local filesystem reads, starts a local network server (port 3456), and accesses SSH credentials.
  • Sanitization: No mention of validation or sanitization of external log content before it is processed or displayed.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
May 17, 2026, 01:36 PM