mgrep-code-search

Warn

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill uses bunx to download the @mixedbread/mgrep package from the NPM registry during execution. This introduces a dependency on external code from a source that is not included in the pre-approved trusted list.
  • [REMOTE_CODE_EXECUTION]: The package downloaded via bunx is executed immediately, allowing for remote code execution from the package source.
  • [INDIRECT_PROMPT_INJECTION]: The skill performs semantic searches over arbitrary codebase content, which presents a surface for indirect prompt injection.
  • Ingestion points: The tool reads and indexes "code, text, PDFs, and images" within the target repository as described in SKILL.md.
  • Boundary markers: The instructions do not define clear boundaries or include warnings to the agent to distinguish search results from system instructions.
  • Capability inventory: The agent can execute shell commands via bunx and access filesystem data.
  • Sanitization: There is no evidence of output sanitization or filtering to prevent the agent from following instructions embedded in the search results.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 15, 2026, 06:05 PM
Security Audit — agent-trust-hub — mgrep-code-search