The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill uses curl to send data, including user prompts and an API key, to https://patentfig.ai. While this domain is the primary service used by the skill, it is not a predefined trusted vendor. Authentication is handled via the PATENTFIG_API_KEY environment variable, which is a secure practice.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for indirect prompt injection by ingesting untrusted data from the user and external image URLs.
Ingestion points: Data enters the agent context through the prompt, referenceImageUrls, and imageUrl parameters in SKILL.md and references/endpoints.md.
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded commands in external data.
Capability inventory: The agent uses shell-executed curl commands to process the input.
Sanitization: The skill lacks explicit instructions for escaping or validating user-provided strings before interpolation into shell command payloads.
[METADATA_POISONING]: The skill metadata lists 'TopLocalAI' as the author, which is inconsistent with the 'davila7' author attribution provided in the analysis context.

patentfig