comet-build
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes task lists (tasks.md), design specifications (design.md), and checkpoint files from the repository to drive its planning and implementation phases. Malicious instructions embedded in these files could potentially influence the agent's behavior.
- Ingestion points: tasks.md, design.md, checkpoint JSON files, and project manifests (package.json, pom.xml, Cargo.toml) are read to determine workflow steps and build commands.
- Boundary markers: The instructions lack specific guidance on using delimiters or ignore-instructions for the ingested repository content.
- Capability inventory: The skill executes shell commands via comet guard and comet check run, modifies persistent state via comet state, and loads external skills via the Skill tool.
- Sanitization: The skill contains explicit instructions to reject shell metacharacters and use direct script execution to mitigate command injection risks.
- [COMMAND_EXECUTION]: The skill executes arbitrary build and test commands provided by the user or found in repository scripts as part of its core build and verification functionality.
- [DYNAMIC_EXECUTION]: The skill performs runtime detection and execution of build scripts defined in project manifest files, which is an intended feature for build automation but constitutes dynamic code execution.
Audit Metadata