research-memory

Pass

Audited by Gen Agent Trust Hub on Nov 9, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because it reads and processes user-generated paper notes from an Obsidian vault to suggest research directions.
  • Ingestion points: The aggregate_reading.py script parses Markdown files from the user's vault, specifically extracting text from 'My Notes' and 'Takeaways' sections.
  • Boundary markers: The agent is provided with note content for summarization without specific escaping or boundary markers beyond structured JSON fields.
  • Capability inventory: The write_research_memory.py script has file-writing capabilities, but these are restricted to the single research-memory.md file within the vault.
  • Sanitization: There is no explicit sanitization of note content; however, the skill's design mandates a multi-round interview process where the user must confirm all proposed changes.
  • Mitigation: The inherent risk is effectively mitigated by the skill's structure, which prevents autonomous writing. A user must review a full unified diff preview before any changes are committed to the file system.
  • [COMMAND_EXECUTION]: Static analysis detected command execution patterns in the test suite (tests/test_research_memory_write.py and tests/test_research_memory_aggregate.py). These are benign developer tools used for verifying the CLI interface of the scripts and are not part of the skill's runtime operations executed by the agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 9, 2026, 09:19 AM