taku-build
Pass
Audited by Gen Agent Trust Hub on May 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill executes shell commands for environment setup and testing based on detected configuration files. Evidence includes the use of
git worktree add,npm install,cargo build,pip install,poetry install,go mod download, and various test runners such asnpm testandpytestwithinreferences/worktrees.md. - [EXTERNAL_DOWNLOADS]: Dependency installation via managers like NPM, Cargo, and Pip involves downloading code from official public registries, which is expected behavior for a build skill.
- [PROMPT_INJECTION]: The skill maintains a surface for indirect prompt injection by ingesting untrusted repository files to drive execution logic. 1. Ingestion points: Reads
PLAN.md,package.json,Cargo.toml,requirements.txt,pyproject.toml, andgo.mod. 2. Boundary markers: Instructions lack markers or specific warnings to ignore instructions embedded in these external files. 3. Capability inventory: Provides capabilities for shell execution (builds/tests) and repository modification (git commits and .gitignore updates). 4. Sanitization: No sanitization or validation of the content within ingested files is performed.
Audit Metadata