The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill documentation describes a process for extracting sensitive session cookies from a running Chrome instance to enable authenticated write access to Figma files.
Evidence: The recon-handshake command is documented to extract cookies and multiplayer URLs from Chrome via the Remote Debugging port and save them to /tmp/figma_handshake.json.
[COMMAND_EXECUTION]: The skill utilizes multiple CLI commands to perform its core functions, including capturing data, decoding binary frames, and writing mutations back to Figma.
Evidence: Documentation includes instructions for using npx figma-kiwi-protocol for various operations like capture-all-pages, decode, and write mutations.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted binary data from Figma's WebSocket protocol, creating a surface for potential indirect instruction injection.
Ingestion points: Intercepts binary WebSocket frames from external Figma sessions via the Chrome DevTools Protocol as described in SKILL.md.
Boundary markers: No evidence of specific boundary markers or delimiters found in the provided documentation to differentiate between data and instructions.
Capability inventory: The skill has extensive capabilities including file-write operations (session.mutate), shell command execution via CLI, and network communication with Figma servers.
Sanitization: There is no mention of sanitization or safety validation for the decoded scenegraph data before it is processed by the agent or used in mutations.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of an external package from a public registry.
Evidence: Instructions provided for npm install figma-kiwi-protocol.

figma-kiwi-protocol-decode