The Agent Skills Directory

[COMMAND_EXECUTION]: The skill contains instructions to execute a specific vendor-owned command-line interface tool for quote confirmation.
Evidence: postplus quote confirm --json --challenge-file <challenge.json>
This command appears to be a legitimate part of the agent's operating environment for managing service quotas or cost approvals.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon external information, including product facts, market evidence, and research data.
Ingestion points: The workflow relies on data inputs during the product_analysis, creator_logic, and sourceBasis phases, often sourced from research skills.
Boundary markers: The skill includes explicit stop conditions if required source evidence or user intent is missing, which serves as a basic control against proceeding with ambiguous or potentially injected data.
Capability inventory: The skill coordinates various media generation runners (image, voice, video) and the postplus CLI tool.
Sanitization: There are no explicit instructions for escaping or sanitizing the input data before it is passed to downstream runners.

ugc-flow