searching-sourcegraph

Pass

Audited by Gen Agent Trust Hub on Jun 19, 2026

Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes content from external repositories. \n
  • Ingestion points: read_file, deepsearch_read, and search results from keyword_search and nls_search (found in SKILL.md and various workflow files). \n
  • Boundary markers: Absent; there are no instructions to the agent to isolate or treat external codebase content as untrusted data. \n
  • Capability inventory: The skill enables deep exploration and reading of code, which could lead an agent to follow malicious instructions embedded in a repository. \n
  • Sanitization: None; raw code content is ingested into the agent context. \n- [DATA_EXFILTRATION]: The skill includes search patterns for identifying sensitive configuration files. In query-patterns.md, it suggests a specific query for environment variables: file:\.env\. [A-Z_]+=. While intended for architectural discovery, this facilitates the location of credentials or secrets that may be improperly stored in a codebase.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 19, 2026, 12:28 PM
Security Audit — agent-trust-hub — searching-sourcegraph