The Agent Skills Directory

[COMMAND_EXECUTION]: Modifies macOS system appearance and accessibility settings to verify application rendering under different conditions.
Evidence: Uses osascript to toggle dark mode and defaults write to adjust settings like increaseContrast, reduceTransparency, and reduceMotion (documented in references/state-matrix.md).
Context: The skill includes a 'harness contract' with snapshot and restoration mechanisms to revert system changes after testing.
[INDIRECT_PROMPT_INJECTION]: The skill processes external, potentially untrusted data sources to drive its automation logic.
Ingestion points: Processes external 'blessed prototype' packages, Regions.md component maps, and SIGNOFF.md sign-off files.
Boundary markers: Includes an explicit policy in references/runtime-trust.md stating that repository files and tool outputs are untrusted data and that embedded instructions cannot alter governing rules or authority.
Capability inventory: Capabilities include system setting modification, application launching, and file writing (PNG/JSON baseline records).
Sanitization: Implements bounded receipts, fixed command interfaces, and 'secret-shaped' name removal from logs.
[DYNAMIC_EXECUTION]: Compiles and runs local tools to perform accessibility audits and interface interaction.
Evidence: Compiles a purpose-built ax-drive Swift tool for Accessibility tree interaction and runs AuditTests.swift via xcodebuild (documented in references/interaction.md).
Context: These are repository-local fixtures used for the specific purpose of automated UI verification.

tailrocks-macos-visual-baseline