The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes store-level data that may originate from third parties (buyers/customers) and includes explicit instructions for the agent to maintain security boundaries.
Ingestion points: Free-text fields in store data including delivery notes, order comments, customer names, product descriptions, and reviews.
Boundary markers: The skill contains a dedicated 'Untrusted store text' section instructing the agent to ignore embedded instructions and continue only with the owner's authorized workflow.
Capability inventory: The agent has access to tools for creating/updating warehouses, uploading files, publishing news articles, and resolving system alerts.
Sanitization: Instructions require the agent to use only minimum excerpts or concise summaries of untrusted text when reporting results to the owner.
[COMMAND_EXECUTION]: The skill utilizes local Node.js scripts for searching documentation and validating request schemas.
Evidence: The allowed-tools field permits Bash(node scripts/search_docs.mjs:*) and Bash(node scripts/validate.mjs:*).
Analysis: The search_docs.mjs script is dependency-free, using only Node.js built-in modules (node:fs, node:url, node:zlib) to read a local gzipped OpenAPI specification file. The skill instructions claim these scripts perform no network operations, which is consistent with the provided source code.

a1-yandex-kit-store