The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines a mechanism to generate and execute short JavaScript snippets through Figma MCP adapters (figma_execute or use_figma). This is used as an 'escape hatch' to retrieve missing metadata from Figma. The instructions strictly limit these scripts to read-only operations and under 50 lines of code, with mandatory auditing in the output JSON.- [SAFE]: The skill operates on local workspace files, specifically reading {cachePath}/{componentSlug}-_base.json and writing to {cachePath}/{componentSlug}-voice.json. It does not access sensitive system credentials or perform unauthorized network exfiltration.- [SAFE]: The 'Batch-mode contract' prevents the agent from pausing for user input or confirmation, which is a standard requirement for orchestrated agent workflows and does not represent a security bypass in this context.- [SAFE]: The skill uses referenced markdown files for accessibility patterns (VoiceOver, TalkBack, ARIA), which are internal workspace references rather than external remote downloads.

extract-voice