inner-life-dream

Pass

Audited by Gen Agent Trust Hub on Mar 6, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The script scripts/should-dream.sh executes a Python subprocess to perform probability calculations. It safely handles the input by using a regular expression (^[0-9]*\.?[0-9]+$) to validate that the configuration value is a decimal number before passing it as a command-line argument, effectively preventing command injection patterns.
  • [PROMPT_INJECTION]: The skill implements logic to scan for 'dream-topic' signals within daily notes, which represents an indirect prompt injection surface.
  • Ingestion points: Markdown files located in memory/daily-notes/ and internal state files like memory/inner-state.json and memory/drive.json.
  • Boundary markers: The script looks for a specific pattern using the regex <!-- dream-topic: \K[^>]+(?= -->) to identify overriding instructions.
  • Capability inventory: The skill allows the agent to write new markdown files to memory/dreams/ and update system-relevant JSON files (inner-state.json, drive.json).
  • Sanitization: While the signal is extracted via a targeted regex, the content within the signal is not sanitized or validated before being presented to the agent as a prompt for creative exploration.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 6, 2026, 03:43 AM