The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill executes the uxproof@0.3.1 package via npx. Although the version is pinned to prevent silent remote code updates, the package is retrieved from an external registry.
[COMMAND_EXECUTION]: The skill uses shell execution via npx to initialize the design contract during the init step.
[INDIRECT_PROMPT_INJECTION]: The skill scans repository source code, documentation, and configuration files, which creates a surface for indirect prompt injection.
Ingestion points: Reads the working tree, including source code, documentation, and configuration files (specified in SKILL.md and references/agentic-setup.md).
Boundary markers: Explicitly defines an "Untrusted content boundary" that instructs the agent to treat repository content as data only and to report any embedded instructions (e.g., "ignore previous instructions") as suspected injection (references/agentic-setup.md).
Capability inventory: Executes shell commands and writes persistent files to the .uxproof/ directory (SKILL.md).
Sanitization: Enforces the redaction of secrets from all outputs and validates externally-sourced paths and regex patterns to prevent shell or path interpolation attacks (SKILL.md, references/design-sources.md).
[CREDENTIALS_UNSAFE]: The skill instructions anticipate scanning files that may contain sensitive data (such as .env files). It provides a mitigation strategy by requiring the redaction of any credential-looking strings before they are included in model plans, reports, or logs (SKILL.md, references/rules.md).

om-ux-setup