The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructions and discovery rubric explicitly direct the agent to access highly sensitive application data paths, including ~/.codex/sessions/, ~/.claude/projects/, and ~/.gemini/. These directories contain detailed interaction logs and transcripts between the user and AI agents, which are highly likely to contain sensitive personal information, project secrets, or credentials. While the skill claims to sanitize data, the broad access to session history constitutes a data exposure risk.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface by ingesting and processing untrusted session transcripts and log files. Malicious instructions embedded in past conversations (e.g., in a code comment or a user's previous request) could potentially influence the agent's behavior during the mining and clustering process. The instructions lack specific prompt boundary markers or input validation to mitigate these risks when reading external files.
[COMMAND_EXECUTION]: The skill relies on an external Python script (scripts/scan_sessions.py) to perform its primary analysis. While the script is intended for local execution, its source code was not provided for audit, making it an unverifiable dependency within the skill package.

skill-miner