skills/poteto/noodle/ruminate/Gen Agent Trust Hub

ruminate

Warn

Audited by Gen Agent Trust Hub on Mar 3, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill accesses and reads sensitive conversation logs from hidden directories in the user's home folder. These files contain a history of user interactions which may include proprietary code, secrets, or personal information.\n
  • Evidence: Accesses ~/.claude/projects/ and ~/.codex/sessions/ to extract interaction data.\n- [COMMAND_EXECUTION]: The skill executes local scripts to perform snapshots and data extraction.\n
  • Evidence: Runs sh .claude/skills/meditate/scripts/snapshot.sh and python3 scripts/extract-conversations.py.\n- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests and processes untrusted data from historical conversation logs.\n
  • Ingestion points: Conversation data is loaded from .jsonl files in ~/.claude/projects/ and ~/.codex/sessions/ using scripts/extract-conversations.py.\n
  • Boundary markers: The extraction script wraps message content in [USER]: and [ASSISTANT]: labels, providing minimal structural separation.\n
  • Capability inventory: The skill has the ability to write to files in the brain/ directory, update SKILL.md files, and modify brain/index.md.\n
  • Sanitization: The script truncates message length but does not sanitize or escape the content to prevent instructions within the logs from being interpreted as commands by the analysis agents.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 3, 2026, 01:34 AM