The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes user-provided source code, which constitutes an indirect prompt injection surface.
Ingestion points: The code parameter in skill.yaml and the generation instructions in SKILL.md.
Boundary markers: Present in SKILL.md as explicit security guardrails instructing the agent to maintain safety boundaries and ignore adversarial pressure to replicate unsafe patterns.
Capability inventory: The skill implementation is limited to string template generation and does not include built-in capabilities for network requests, file writes, or command execution.
Sanitization: Mandatory instructions require the agent to use placeholders for secrets and replace dynamic code execution with static tests.
[DATA_EXFILTRATION]: The skill includes specific guardrails to prevent the leakage of sensitive data by instructing the agent to mock credentials, secrets, and connection strings found in source code with placeholder values. This prevents production secrets from being committed to version control in test files.
[EXTERNAL_DOWNLOADS]: The skill provides instructions that explicitly forbid the installation of new packages or the modification of project configuration files (like package.json or requirements.txt) without explicit user consent, mitigating supply chain risks associated with automatic dependency management.
[SAFE]: The implementation in skill.yaml uses a safe templating approach, automatically prepending parameter names with 'mock' to ensure generated code remains isolated from real data inputs.

test-case-generator-hardened