The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes user-defined workflow descriptions to generate instructions for new skills, creating an attack surface for indirect prompt injection where malicious user input could be incorporated into the resulting skill code.
Ingestion points: User-provided descriptions of workflows, use cases, and triggers during the Discovery and Architecture phases (SKILL.md).
Boundary markers: No delimiters or "ignore embedded instructions" warnings are utilized when the agent interpolates user input into instructions.
Capability inventory: The skill uses present_files to create and write instructions to a new SKILL.md file.
Sanitization: No validation or sanitization of the user-provided workflow content is performed before drafting the instructions.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute a provided Python script (scripts/validate_skill.py) to perform static analysis on generated skill files. This script operates locally and does not perform network operations or execute arbitrary shell commands.
[SAFE]: The author name 'Felipe Rodrigues' in the skill metadata differs from the platform-reported author 'marcioaltoe'. This inconsistency is documented neutrally as it likely reflects template usage or distribution rather than a malicious deception attempt.

skill-architect