The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes functionality that processes untrusted external data, creating a potential attack surface for indirect prompt injection.
Ingestion points: The skill mentions acquiring video, reading captions, and running local transcription and OCR (SKILL.md).
Boundary markers: The instructions do not specify the use of delimiters or boundary markers to isolate processed data from the system instructions.
Capability inventory: The skill is configured to use Bash and Read tools, which could be targeted if the processed data contains malicious instructions.
Sanitization: No specific sanitization or filtering logic is described in the configuration guide for handling external content.
[COMMAND_EXECUTION]: The skill provides examples for using the watch-skill CLI tool to configure vision providers. This tool is a vendor-specific resource associated with the skill's purpose.
[CREDENTIALS_SAFE]: The skill demonstrates safe credential management practices by using <KEY> placeholders in examples and explicitly instructing the agent not to ask users to reveal secrets in chat, recommending environment variables instead.

configuring-vision