The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements features for auditing and analyzing external web content (HTML and URLs), which exposes the agent to indirect prompt injection risks if the processed data contains malicious instructions designed to manipulate the agent's behavior.
Ingestion points: The analyzeContent method and the audit workflow ingest untrusted HTML content and URLs for analysis.
Boundary markers: The instructions do not define clear delimiters or specify "ignore instructions" directives when processing the ingested content.
Capability inventory: The skill instructions recommend "improving" and "fixing" site structure, implying the agent may use tools to modify files or configurations based on the audit results.
Sanitization: There is no mention of sanitization, filtering, or validation of the external content before it is processed by the agent.
[EXTERNAL_DOWNLOADS]: The documentation references an external GitHub repository (github.com/ericosiu/ai-marketing-skills) as a source for its SEO methodology. This points the agent toward unverified logic from a source not recognized as a trusted organization or well-known service.
[OBFUSCATION]: The .clawnet/unsigned-skill.json file contains a hex-encoded payload in the opReturnHex field. Decoding reveals that this is a representation of the skill's own metadata and documentation, likely used for blockchain-anchored integrity verification, but it qualifies as encoded content that obscures plain-text analysis.

ai-seo-optimization