The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Figma nodes and reference implementations to generate defect reports and influence UI code modifications.
Ingestion points: SKILL.md defines primary inputs as "Figma node(s) via the Figma MCP companion" and "reference screens / a reference implementation."
Boundary markers: The instructions do not specify the use of delimiters or specific "ignore embedded instructions" warnings for the content retrieved from external Figma nodes.
Capability inventory: The skill is authorized to write defect lists and inventories to the file system (in <artifacts-root>/specs/pixel-audit/), modify UI library component source code, and execute JavaScript within a browser environment.
Sanitization: The instructions lack guidance on sanitizing or validating metadata and variables retrieved from Figma before they are used in automated processes or code changes.
[COMMAND_EXECUTION]: The skill utilizes shell commands and network utilities for environment verification and task distribution.
references/evidence-capture.md directs the agent to use curl to inspect served assets and verify that build pipeline changes have propagated to the runtime.
SKILL.md instructs the agent to dispatch local sub-agents ("lanes") to handle mapping tasks independently.
[DYNAMIC_EXECUTION]: The skill uses browser automation to execute arbitrary JavaScript for UI inspection.
references/evidence-capture.md specifies the use of browser evaluation to call el.getBoundingClientRect() and getComputedStyle(el) for gathering precise element metrics as proof for the verification gate.

pixel-audit