symdex-code-search
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill functions by indexing and retrieving content from external repositories (source code, documentation, and markdown files). This establishes a surface for indirect prompt injection, where malicious instructions hidden in a repository's code or comments could attempt to influence the agent's reasoning when the code is searched or retrieved.
- Ingestion points: The skill ingests data from local folders or remote repositories via
index_folderandindex_repo. - Boundary markers: While the skill retrieves specific symbols or file outlines to limit context, it does not explicitly define mandatory delimiters to isolate untrusted repository content from instructions.
- Capability inventory: The skill possesses capabilities for repository indexing, file reading, and searching, but restricts these to structural and semantic analysis within the workspace.
- Sanitization: The instructions emphasize inspecting
qualitymetadata (confidence, freshness, parser mode) before reasoning from results, which serves as a manual verification step for the agent.
Audit Metadata