codebase-management

Pass

Audited by Gen Agent Trust Hub on Sep 18, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill indexes source code and documentation from project directories into a vector database. This creates a surface where external data is ingested into the agent's context.\n
  • Ingestion points: Any files within the specified projectPath and artifacts defined in .socraticodecontextartifacts.json (SQL, OpenAPI, Markdown, etc.).\n
  • Capability inventory: Tools like codebase_index and codebase_graph_build process these files to create embeddings and dependency graphs.\n
  • Boundary markers: The management instructions do not specify prompt-level delimiters for indexed content.\n
  • Sanitization: Content is chunked and processed using embedding models for future retrieval-augmented generation.\n- [COMMAND_EXECUTION]: The management tools coordinate background indexing tasks and file watchers. The codebase_stop tool has the capability to send SIGTERM to orphan processes holding file locks to resolve cross-process conflicts.\n- [CREDENTIALS_UNSAFE]: The documentation describes how to configure API keys for OpenAI, Google, and Qdrant. It recommends using environment variables for secret management and does not contain hardcoded credentials.\n- [EXTERNAL_DOWNLOADS]: Initial setup involves downloading infrastructure components like Docker images and embedding models. The documentation points to well-known services and official repositories for these resources.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 18, 2026, 11:02 AM
Security Audit — agent-trust-hub — codebase-management