The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill incorporates templates and logic to use the com.apple.security.temporary-exception.files.absolute-path.read-write entitlement. This is explicitly documented in the changelog as a 'helper app pattern to bypass sandbox restrictions,' allowing the generated extensions to access system files outside the standard application sandbox.
[DYNAMIC_EXECUTION]: The Swift templates in references/applescript-iterm.swift use NSAppleScript to execute command strings constructed at runtime. This allows the execution of arbitrary AppleScript, including opening terminal sessions and executing directory changes.
[INDIRECT_PROMPT_INJECTION]: The skill processes natural language requests to generate build configurations (xcodegen-template.yml) and Swift source code (finder-sync-template.swift). Maliciously crafted user input could potentially influence the resulting logic or permissions of the generated binaries.
Ingestion points: Natural language requests in SKILL.md and user profile data via scripts/profile_store.py.
Boundary markers: The skill relies on natural language boundary descriptions but lacks strict character-level escaping for code generation.
Capability inventory: File system write access for project creation and CLI execution for building/registering extensions (xcodebuild, xcodegen, pluginkit).
Sanitization: scripts/profile_store.py implements a filter to prevent storing keys containing sensitive strings like 'token', 'secret', or 'password'.
[COMMAND_EXECUTION]: The skill requires the execution of multiple local system utilities and Python scripts to function, including building macOS binaries and validating skill structures.

lov-finder-action