coin-flip

Pass

Audited by Gen Agent Trust Hub on May 12, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes historical conversation data which could contain instructions intended to influence the agent's behavior (indirect prompt injection). 1. Ingestion points: Reads conversation logs from ~/.claude/projects/ (SKILL.md Step 1 and 2). 2. Boundary markers: No explicit delimiters or instructions are provided to distinguish between historical data and potential embedded commands. 3. Capability inventory: The skill is focused on text analysis and does not utilize tools for network access, file modification, or command execution beyond reading logs. 4. Sanitization: No specific sanitization or filtering of historical content is mentioned.
  • [COMMAND_EXECUTION]: The skill uses the find command to identify recent log files within the ~/.claude/projects/ directory. This is a standard operation necessary to gather context for the analysis.
Audit Metadata
Risk Level
SAFE
Analyzed
May 12, 2026, 12:58 PM