The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes allowed Bash tools to manage local files and execute a dedicated Python validation script.
Evidence:
In steps/create.md and steps/revise.md, the skill executes python3 "$SKILL_DIR"/scripts/validate.py to check the validity of generated prompts.
It uses mkdir, cp, and mv commands to organize project directories (tracks/, versions/) and archive previous iterations of TRACK.md files.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted user data in the form of music briefs and feedback to synthesize prompt artifacts.
Ingestion points: The create subcommand takes a <brief> argument, and the revise subcommand takes a <feedback> argument, both of which are free-text inputs provided by the user in the $ARGUMENTS field of SKILL.md.
Boundary markers: The generated prompts are encapsulated within TRACK.md using triple-backtick fences (e.g., ```text ... ```) to separate the prompt content from the rest of the documentation.
Capability inventory: The skill possesses the ability to write files to the local file system and execute the local scripts/validate.py script using the Python 3 interpreter via the Bash tool.
Sanitization: The skill employs a deterministic linter (scripts/validate.py) that strictly enforces character limits and utilizes regular expressions to detect and block artist citation patterns (e.g., "in the style of..."). This prevents the generation of prompts that might trigger safety filters or create legal risks.

suno-produce