The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes project source code to discover application views and routes, creating a surface where untrusted data can influence the agent's automation logic.
Ingestion points: Project files including Swift/SwiftUI structs, web routes, and Electron configuration (SKILL.md, Step 1).
Boundary markers: No explicit delimiters or instructions to ignore embedded content are used when the agent reads external code files.
Capability inventory: The skill utilizes shell command execution (screencapture), UI automation (osascript), and browser automation (Puppeteer/Playwright).
Sanitization: Discovered view names and routes are used to generate filenames and navigate automated browsers without specified sanitization steps.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands and AppleScript to interact with the host system and manage the file system.
Evidence: Use of osascript to query system window IDs and activate processes, and the screencapture utility to write image files to disk.
[DYNAMIC_EXECUTION]: The skill generates and runs browser automation scripts (Puppeteer or Playwright) at runtime based on the routes discovered during project analysis.

app-screenshots