skills/letta-ai/skills/memfs-search/Gen Agent Trust Hub

memfs-search

Warn

Audited by Gen Agent Trust Hub on Apr 3, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill requires the installation of the '@tobilu/qmd' package from the public NPM registry. It also fetches approximately 2GB of embedding models from Hugging Face during its setup process.
  • [REMOTE_CODE_EXECUTION]: By instructing users to install and run the third-party '@tobilu/qmd' package, which is not authored by a recognized trusted organization, the skill introduces a dependency on external code that could pose a risk if the package or its registry entry is compromised.
  • [COMMAND_EXECUTION]: The skill uses a bash script ('scripts/memfs-search.sh') as a wrapper for the 'qmd' tool. This script executes various shell commands to manage memory collections, generate embeddings, and perform queries.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It ingests untrusted data from the agent's memory directory ('$MEMORY_DIR'), and the resulting search snippets are returned to the agent without sanitization or boundary markers, potentially allowing malicious instructions in memory files to influence agent behavior.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 3, 2026, 04:05 AM