readwise-rival

Pass

Audited by Gen Agent Trust Hub on Mar 6, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes python3 -c to run embedded Python logic for file operations and data searching within the ~/.xiaodazi/ directory.
  • [PROMPT_INJECTION]: Potential for indirect prompt injection as the skill ingests and processes untrusted text from external sources (highlights).
  • Ingestion points: External text (highlights) is stored in JSON files under ~/.xiaodazi/reading/highlights/.
  • Boundary markers: No explicit delimiters or instructions are used to separate untrusted data from agent instructions during retrieval or summarization.
  • Capability inventory: Includes file system access (read/write) and shell command execution (python3 -c).
  • Sanitization: The skill lacks validation or sanitization for the content of highlights before they are re-processed by the LLM for card generation.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 6, 2026, 04:20 PM