lean-memory

Pass

Audited by Gen Agent Trust Hub on Nov 21, 2026

Risk Level: SAFECOMMAND_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The script scripts/mem.py utilizes subprocess.run to execute Git commands (init, add, commit, push, pull, status, etc.) for repository management. These commands are necessary for the synchronization of the memory store across different environments and to detect local repository slugs.
  • [PERSISTENCE]: During the setup process, the skill modifies the configuration of Claude Code (settings.json) to install a SessionStart hook. This hook automatically runs python3 mem.py context when a new session begins, ensuring shared memory is consistently loaded.
  • [INDIRECT_PROMPT_INJECTION]: The skill facilitates the injection of instructions and facts fetched from a remote Git repository into the agent's context, creating an attack surface for indirect prompt injection.
  • Ingestion points: Markdown entries stored in the ~/.agent-memory directory, which are pulled from a user-configured Git remote.
  • Boundary markers: Injected context is prefaced with an # Agent memory header and a disclaimer instructing the model to treat the content as data rather than absolute instructions.
  • Capability inventory: The skill can modify agent configuration files (e.g., settings.json), create filesystem symlinks to redirect instruction files, and perform network operations via Git subprocesses.
  • Sanitization: The script includes a SECRET regex scanner that checks memory entries for common sensitive patterns (e.g., private keys, API tokens) and blocks their addition to the store.
  • [DATA_EXFILTRATION]: The skill is designed to sync local memory entries to a remote Git URL. While intended for legitimate user-controlled synchronization, this mechanism provides a channel through which sensitive preferences or environment data could be transmitted to an external server.
  • [EXTERNAL_DOWNLOADS]: The documentation in references/agent-setup.md provides instructions to download and install the skill from a remote repository using npx skills add aussadawut-dev/lean-skills.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 21, 2026, 01:13 AM