mining-safety
Warn
Audited by Gen Agent Trust Hub on Mar 23, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill contains a 'Self-Evolution Telemetry' section that instructs the agent to search for and write to paths within
~/.claude/projects/. This directory is a sensitive system path for the Claude Code agent, containing project-specific metadata, history, and internal state. Writing to these files is a persistence mechanism that bypasses standard project boundaries and could expose or corrupt internal agent data. - Evidence: 'Look for the project path in ~/.claude/projects/' and 'append to skill-telemetry.md in that memory directory' found in the SELF-EVOLUTION TELEMETRY section.
- [PROMPT_INJECTION]: The skill processes untrusted data from the local codebase and user-supplied arguments without implementing security best practices to prevent indirect prompt injection. Malicious instructions embedded in incident reports, safety manuals, or code comments could influence agent behavior.
- Ingestion points: User arguments ($ARGUMENTS) and the 'actual codebase' mentioned in the instructions.
- Boundary markers: Absent. The skill does not use delimiters or explicit instructions to treat external data as non-executable text.
- Capability inventory: The skill utilizes file system access to read project data and write telemetry to internal agent directories.
- Sanitization: Absent. There is no logic provided to filter or escape content ingested from the codebase or arguments.
Audit Metadata