The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides instructions for the agent to process external data sources such as website copy, customer surveys, and chat logs, which introduces a surface for indirect prompt injection.
Ingestion points: Analyzing website landing pages, customer surveys, chat logs, support tickets, and reviews as described in SKILL.md and references/RESEARCH.md.
Boundary markers: Absent; the skill does not specify delimiters to separate untrusted analysis data from the agent's instructions.
Capability inventory: The skill is primarily instructional and does not perform tool calls or file writes itself, but relies on an agent's existing file-reading capabilities.
Sanitization: Absent; no validation or sanitization steps are defined for processing external feedback content.
[EXTERNAL_DOWNLOADS]: The skill includes a referral link to an external book resource on Amazon.
Evidence: Affiliate link for 'Making Websites Win' in SKILL.md.
[METADATA_POISONING]: A discrepancy exists between the author name in the skill metadata ('wondelai') and the provided author context ('manish31-oss'), although this appears to be a documentation inconsistency rather than a malicious deception.

cro-methodology