The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses phrases like 'IMPORTANT', 'CRITICAL', and 'Ignore' in the context of its own internal methodology (design-by-contract and lean-startup practices). These are standard instructional terms and do not attempt to bypass agent safety filters or override system-level instructions.
[COMMAND_EXECUTION]: The skill includes a scripts/validate.sh file that executes a script (heal.sh) from the skills/skill-builder directory. This is a local repository reference for internal skill validation and does not invoke arbitrary or untrusted shell commands.
[DATA_EXFILTRATION]: No network operations or sensitive file access patterns were detected. The skill operates on goal descriptions and architectural design patterns supplied by the user.
[EXTERNAL_DOWNLOADS]: The skill does not perform any external network requests or download third-party code. All references are local to the skill's file structure.
[REMOTE_CODE_EXECUTION]: The skill does not contain any remote code execution patterns. It explicitly mandates 'zero goal executions' and 'stop after 1 lint pass,' ensuring it only processes text without running the generated goal prompts.
[CREDENTIALS_UNSAFE]: No hardcoded credentials, API keys, or secrets were found in the skill or its associated references.
[NO_CODE]: The skill primarily consists of markdown instructions and a template. The only executable script is a validation utility for development, which is standard for the environment.

craft-goal