The Agent Skills Directory

[SAFE]: The skill is entirely instructional, providing prompt templates and guidelines for using an image generation tool. It does not include any executable scripts, network requests, or file system modifications beyond saving the generated image to a user-defined path.
[INDIRECT_PROMPT_INJECTION]: The skill identifies a potential attack surface through the ingestion of user-provided photos.
Ingestion points: The INPUT IMAGE path in the prompt template in SKILL.md.
Boundary markers: The template uses specific markdown headers (## INPUT IMAGE, ## CRITICAL: LIKENESS PRESERVATION) and explicit negative constraints (USE ONLY THIS IMAGE) to isolate the input.
Capability inventory: The skill uses image generation capabilities and standard file-write operations for the output.
Sanitization: The skill instructs the agent to focus on specific facial features extracted from the image, which acts as a form of descriptive sanitization by ignoring non-feature data.
[METADATA_POISONING]: The .clawnet/unsigned-skill.json file contains a large opReturnHex field. Analysis confirms this is a standard Metanet protocol encoding of the skill's own markdown content for blockchain-based verification and does not contain hidden commands or malicious payloads.

avatar-portrait