search-context
Pass
Audited by Gen Agent Trust Hub on Aug 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill interacts with the local environment by executing standard command-line utilities such as git, gh (GitHub CLI), and rg (ripgrep) via Node.js spawnSync. The implementation correctly passes arguments as arrays, avoiding shell interpolation and effectively preventing command injection vulnerabilities.- [EXTERNAL_DOWNLOADS]: The skill performs shallow, blobless clones of public GitHub repositories. This is the core intended functionality and is implemented using security-conscious flags (--depth=1 --filter=blob:none) to minimize the amount of data processed from external sources.- [INDIRECT_PROMPT_INJECTION]: The skill extracts code snippets from untrusted third-party repositories to provide context to the AI agent. While this exposes a potential injection surface, the risk is mitigated by explicit instructions and boundary markers.
- Ingestion points: The scripts/search-context.mjs file uses ripgrep to read content from cloned repository directories.
- Boundary markers: The generated Markdown manifest follows a template (templates/reference-manifest.md) that includes specific agent instructions to treat the data as reference material and avoid copying large blocks.
- Capability inventory: The skill has no capabilities for executing the retrieved code, installing dependencies, or making arbitrary network connections beyond GitHub API and Git operations.
- Sanitization: Code snippets are encapsulated in Markdown blocks, and the skill includes documentation (references/security.md) that explicitly forbids the execution of scripts or binaries from reference repositories.
Audit Metadata