mem-recall
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: Instructions recommend executing shell commands using the
trellis memCLI (e.g.,trellis mem projects,trellis mem search,trellis mem context, andtrellis mem extract) to interact with local databases. - [EXTERNAL_DOWNLOADS]: The prerequisite section guides the user or agent to install the global npm package
@mindfoldhq/trellis@latestfrom the official npm registry. - [INDIRECT_PROMPT_INJECTION]:
- Ingestion points: Reads external local data from historical AI session files across multiple platforms (
~/.claude/projects/,~/.local/share/devin/cli/sessions.db, etc.) into the active agent context. - Boundary markers: The tool specifies structural cleaning rules, filtering out known platform system tokens (e.g.,
<system-reminder>,<workflow-state>). - Capability inventory: Uses subprocess command execution (
trellis mem) to query local file structures and SQLite databases. - Sanitization: Stated to perform automated stripping of tool noise and hook injections from raw logs before exposing context to the model.
Audit Metadata