The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use ffmpeg and ffprobe for reading source-frame metadata and performing non-editorial diagnostics on local files. It also utilizes curl to download temporary image resources from signed URLs for visual verification when hosted assets are not directly inspectable.
[EXTERNAL_DOWNLOADS]: Instructions are provided to register a remote MCP (Model Context Protocol) server using codex mcp add chatcut --url https://api.chatcut.io/api/external-mcp/mcp. This allows the agent to download and interface with external tools hosted on the vendor's infrastructure.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from external sources, including video transcripts, asset metadata, and timeline layouts, which could potentially contain malicious instructions.
Ingestion points: Untrusted content enters the agent's context through tools such as browse_assets, inspect_asset, and preview_timeline described in SKILL.md.
Boundary markers: There are no explicit instructions for the agent to use delimiters or ignore embedded commands within the processed transcripts.
Capability inventory: The agent has the ability to execute shell commands (ffmpeg, curl) and perform wide-ranging project modifications via the mcp__chatcut__* toolset.
Sanitization: The skill does not define methods for sanitizing or validating the content of transcripts or metadata before it is interpolated into prompts.

chatcut-plugin-basics