The Agent Skills Directory

[COMMAND_EXECUTION]: The documentation includes shell commands for project environment setup, such as setting environment variables (export SKIP_SKILL_GUARDRAILS), modifying file permissions for hook scripts (chmod +x), and performing local testing of TypeScript hooks using npx tsx and jq.
[INDIRECT_PROMPT_INJECTION]: The skill documents an architecture that processes untrusted data (user prompts and file content) to trigger automated context injection and tool enforcement.
Ingestion points: The UserPromptSubmit and PreToolUse hooks read prompt data and file contents through stdin, as detailed in HOOK_MECHANISMS.md.
Boundary markers: SKILL.md identifies the // @skip-validation marker as a mechanism to bypass automated enforcement.
Capability inventory: The system described has the capability to inject formatted text into the agent's context and block the execution of critical tools like Edit and Write.
Sanitization: The documentation describes regex-based matching but does not explicitly detail input sanitization or escaping protocols for the processed data.

skill-developer