The Agent Skills Directory

[METADATA_POISONING]: The skill instructions contain significant internal contradictions; while the initial name and description focus on 'Audio Stem Separation', the subsequent 'Step 4' and 'Inputs' sections describe image generation, editing, and restyling functionality. This misleading metadata may cause the agent to misinterpret the skill's actual capabilities.
[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection by processing external audio URLs and user-supplied prompts which are then sent to a remote API gateway.
Ingestion points: Untrusted data enters via the audio_url parameter and the natural language prompt provided by the user (SKILL.md).
Boundary markers: No boundary markers or 'ignore embedded instructions' warnings are defined to delimit external data from agent instructions.
Capability inventory: The skill utilizes network capabilities via curl, Python requests, and Node.js fetch to interact with remote endpoints.
Sanitization: There is no evidence of data sanitization, validation, or escaping for user-supplied content before it is interpolated into the API requests.
[EXTERNAL_DOWNLOADS]: The skill is configured to fetch technical reference material and model schemas from the vendor's official domain (www.pixazo.ai) to resolve parameter requirements.
[COMMAND_EXECUTION]: The documentation includes a command for the user to install additional vendor tools via npx from the Pixazo-AI/skills repository.

separate-stems-api