The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from Figma designs, including layer names, frame titles, and text content, which are processed during the handoff workflow steps.\n- Ingestion points: Figma metadata, node properties, and layer content are read via the Figma API or helper tools as described in SKILL.md (Workflow Step 2 and 3).\n- Boundary markers: No explicit instruction delimiters or 'ignore embedded instructions' warnings were found in the skill's reference materials or scripts to prevent the agent from obeying instructions hidden in design data.\n- Capability inventory: The skill is configured with high-privilege tools including Bash, Read, Write, and Edit, as listed in the SKILL.md frontmatter.\n- Sanitization: The scripts/theme-figma.js script performs basic sanitization and normalization using regex for node IDs and section names, but does not sanitize content against instruction injection.\n- [COMMAND_EXECUTION]: The skill's test suite in tests/test_asset_contract.py utilizes subprocess.run to execute Python and Node.js validation scripts.\n- Evidence: The tests invoke local script files like scripts/theme-figma.js to verify handoff package integrity and schema compliance.\n- [DYNAMIC_EXECUTION]: The unit tests include a dynamic execution pattern to introspect the skill's own helper script for validation purposes.\n- Evidence: A test probe in tests/test_asset_contract.py uses node -e to dynamically require the skill's helper script by its file path to check internal constant mappings.

next-theme-figma