claude-md-template
Pass
Audited by Gen Agent Trust Hub on Aug 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions in Step 5 direct the agent to verify generated commands by executing them in the local environment, including installation and build tasks.
- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection by ingesting data from repository files to generate commands that are then executed by the agent.
- Ingestion points: The agent reads project configuration files such as package.json, pyproject.toml, and existing CLAUDE.md files (SKILL.md, Steps 1 and 2).
- Boundary markers: The template includes 'Guardrails' and 'Security boundaries' sections which provide natural language instructions for safe operation, though these do not prevent adversarial exploitation of the command verification step.
- Capability inventory: The agent performs shell command execution and file system write operations.
- Sanitization: No explicit sanitization or validation of commands derived from external configuration files is performed before the verification execution.
Audit Metadata