The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external artifacts, including AI-generated UI from various platforms (v0.dev, Galileo, Builder.io) and specification files like ux-flow-spec.json and ui-component-spec.json.
Ingestion points: SKILL.md identifies multiple data entry points including feature tickets, flow specs, component specs, and research reports.
Boundary markers: While specific textual delimiters are not defined, the skill provides instructions to cross-check copy against declared goals.
Capability inventory: The skill environment (as restricted in frontmatter) includes write_file, run_tests, and execute_command capabilities.
Sanitization: The skill explicitly includes a guardrail (ASI05 RCE Guard) instructing the agent to never embed user-supplied content without sanitization, significantly mitigating the risk.
[SAFE]: No malicious code, obfuscation, data exfiltration, or unauthorized privilege escalation patterns were found. The skill's permissions and tool usage are consistent with its stated purpose of performing accessibility scans (axe-core) and managing design-to-code pipelines (Style Dictionary).

design-review