The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions require the agent to ingest and analyze data from external sources, including product pricing pages, checkout screens, and confirmation emails, which creates a surface for indirect prompt injection.
Ingestion points: Data enters the agent's context from third-party websites during the inspection of pricing, terms, and purchase flows as described in SKILL.md.
Boundary markers: The skill mandates the use of specific evidence labels (such as OBSERVED, SOURCE-INDICATED, and DECLARED) to categorize findings, which serves as a prompt-level boundary to differentiate between external data and internal instructions.
Capability inventory: The skill is authorized to navigate web pages, exercise checkout flows in test mode, and reload pages to verify timers. It does not contain instructions for sensitive file access, administrative shell execution, or unauthorized network exfiltration.
Sanitization: There are no explicit instructions for technical sanitization or escaping of the external content processed during the audit.
[EXTERNAL_DOWNLOADS]: The skill provides a comprehensive list of official regulatory and standards-body websites (e.g., eur-lex.europa.eu, ftc.gov, ico.org.uk, w3.org) in references/official-sources.md for the agent to reference legal requirements and guidance. These are well-known official government and international organizations used to support the audit process.

consumer-protection-readiness