The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses bun to run local automation scripts (run.ts, install.ts), xcodebuild for test compilation, and osascript or defaults write to modify macOS system appearance and accessibility settings. These commands are fundamental to the skill's primary purpose of visual QA and are managed through bounded transactions that require restoration to the original system state.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data such as repository fixtures, scripts, and tool outputs which are explicitly categorized as untrusted.
Ingestion points: Ingests files from the repository root, screen/state matrices, and output from AX driving tools (identified in SKILL.md and runtime-trust.md).
Boundary markers: Employs mandatory preflight checks for permissions and byte-binds the harness to recorded sources to prevent unauthorized modification.
Capability inventory: Orchestrates shell execution for bun, xcodebuild, ax-drive, and system configuration utilities.
Sanitization: runtime-trust.md establishes a policy that embedded instructions in processed data cannot alter agent scope or authority, and sensitive values are excluded from logs and prompts.

tailrocks-macos-visual-qa