The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes internal business artifacts and web search results to generate design system configurations and HTML documentation.\n
Ingestion points: The agent reads empathy maps, personas, and other UX artifacts from the biz/ directory and fetches external data using the WebSearch and WebFetch tools as described in SKILL.md.\n
Capability inventory: The skill has the capability to write local files and execute shell commands to open them.\n
Sanitization: There is no evidence of explicit content sanitization for data retrieved from web sources before it is used in the generation process.\n
Boundary markers: As a mitigation, Step 8 in SKILL.md requires a human-in-the-loop approval of the derived brand strategy before finalizing outputs.\n- [DATA_EXFILTRATION]: Internal business context is used to generate and execute external web search queries.\n
Evidence: Step 4 in SKILL.md describes generating search queries based on archetype, tech-comfort, and industry data extracted from files in the biz/ directory.\n
Impact: This behavior transmits project-specific context and audience details to external search engine providers.\n- [COMMAND_EXECUTION]: The skill uses shell commands to display its outputs to the user.\n
Evidence: SKILL.md specifies using xdg-open or open to preview the generated brand-sheet-{subject}.html file.\n- [EXTERNAL_DOWNLOADS]: The skill fetches content from external websites for benchmarking and reference research.\n
Evidence: The skill uses WebFetch to visit reference sites. A curated fallback list in scripts/benchmark-defaults.yaml includes well-known technology and institutional services such as Stripe, GitHub, Vercel, and various French government portals.

ux-brand