The Agent Skills Directory

[COMMAND_EXECUTION]: The references/checks.md file contains shell commands using curl, grep, and sed to retrieve and process ecommerce policy information from user-provided storefront URLs. These tools are used for their intended purpose of content extraction.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external websites, which represents an indirect prompt injection surface. The risk is mitigated by explicit safety instructions.
Ingestion points: Untrusted content enters the agent's context through curl operations defined in references/checks.md targeting external storefront URLs.
Boundary markers: SKILL.md implements guardrails requiring the agent to ground all answers in verified policy sources and to use HOLD or BLOCK states when information is unverifiable.
Capability inventory: The skill utilizes shell execution for networking (curl) and text processing (sed, grep, tr, head).
Sanitization: The references/checks.md script includes basic HTML stripping logic (sed 's/<[^>]*>/ /g') to clean the ingested text before the agent processes it.

ecommerce-policy-readiness