codegraph-ast-grep
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads binaries and release assets for CodeGraph and ast-grep from GitHub repositories. These operations are performed with strict verification steps, including SHA-256 checksum checks and archive inspection to prevent the execution of malicious or corrupted assets.
- [COMMAND_EXECUTION]: Orchestrates tool installation using standard package managers like npm, pnpm, cargo, and pip. It emphasizes security by using local project shims and absolute paths for execution, which prevents dependency shadowing and ensures that only reviewed, project-specific binaries are invoked.
- [INDIRECT_PROMPT_INJECTION]: The skill analyzes repository source code to provide semantic and structural intelligence, which presents a surface for indirect prompt injection.
- Ingestion points: Static analysis of repository files during indexing and search workflows (SKILL.md, references/usage-playbook.md).
- Boundary markers: Persists clear repository-native guidance to agents regarding tool capabilities and usage rules.
- Capability inventory: Executes external CLI tools and writes configuration files (codegraph.json, sgconfig.yml) within the repository root.
- Sanitization: Instructions mandate the redaction of secrets, private identifiers, and customer data from all generated artifacts.
- [DYNAMIC_EXECUTION]: Generates shell and PowerShell shims at runtime to ensure tools are executed with appropriate environment variables, such as disabling telemetry and binding to the correct project root.
- [SAFE]: The skill adheres to strong privacy principles by defaulting telemetry to off (CODEGRAPH_TELEMETRY=0), respecting DO_NOT_TRACK signals, and requiring explicit human-in-the-loop approval before performing metadata migrations or index repairs.
Audit Metadata