The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions in SKILL.md direct users to add components using bunx --bun shadcn@latest add Boring-Software-Inc/dither-kit/<component>. This command fetches and executes code from a repository belonging to an unverified third-party entity (Boring-Software-Inc), posing a risk of supply chain attack or execution of untrusted code.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to generate UI specifications from external data sources such as user preferences and application state (as seen in README.md and SKILL.md). This ingestion of untrusted data into the agent's generation pipeline represents an indirect prompt injection surface. While the skill emphasizes the use of Zod schemas for structural validation, it lacks explicit boundary markers or instructions to ignore potential commands embedded within the data fields (e.g., titles or labels) that the agent processes.
[OBFUSCATION]: The .clawnet/unsigned-skill.json file contains a large hex-encoded string in the opReturnHex field. Analysis indicates this encodes the skill's own instructions and metadata, likely for blockchain-based verification using the B-protocol prefix 19HxigV4QyBv3tHpQVcUEQyq1pzZVdoAutM. Although consistent with certain platform standards, it constitutes a form of content encoding that hides instructions from plain-text inspection.

generative-ui