The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external product imports and storefront URL inspections.
Ingestion points: The catalog import preview command (documented in references/bulk-import.md) processes external product lists from various sources, and the storefront review (documented in references/audit-and-recommendations.md) processes content from public URLs.
Boundary markers: SKILL.md and references/images.md include explicit instructions to treat all external content (product names, descriptions, scraped pages, URLs) as untrusted data and never as instructions.
Capability inventory: The skill has the capability to perform network operations and modify catalog data, although state-changing actions are constrained by the approval workflow.
Sanitization: All mutations are gated by a durable server-side proposal and human-in-the-loop approval process, where the agent cannot apply changes without a human-verified approval ID.
[EXTERNAL_DOWNLOADS]: The skill uses npx --yes to fetch and execute specific versions of the @tarileo/avileo-catalog-cli package.
[COMMAND_EXECUTION]: The skill invokes shell commands using the Avileo CLI and provides instructions for local development workflows that involve using bun to build the SDK and execute code from source directories.

avileo-catalog