light-memory
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external research data (papers, notes, conversation excerpts) which could contain instructions. However, it implements robust mitigations:\n
- Ingestion points: Untrusted data enters the agent context via research files and excerpts processed in
memory_store.py.\n - Boundary markers: The skill places explicit warnings and delimiters in its generated views (
current.md) and limits character budgets to prevent prompt flooding.\n - Capability inventory: The skill is limited to reading and writing project files; it does not have network access or the ability to execute arbitrary commands based on ingested data.\n
- Sanitization: It performs NFKC normalization and strict path validation before processing any file-based evidence.\n- [COMMAND_EXECUTION]: The test suite (
tests/test_memory.py) usessubprocess.runto verify the CLI functionality. These calls are limited to the skill's own scripts and do not involve user-controlled shell execution.\n- [SAFE]: The code follows security best practices, including atomic file operations, strict project-scoping for file access, and preventing the ingestion of sensitive files such as credentials or repository metadata.
Audit Metadata