The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The Workflows/VmLifecycle.md file includes hardcoded default credentials (lume) within example commands intended for migrating from external virtual machine management tools.
[PRIVILEGE_ESCALATION]: The uninstallation and update workflows in Workflows/Update.md utilize sudo to delete binary files and frameworks from system-protected directories, including /usr/local/bin and /usr/local/Frameworks.
[DYNAMIC_EXECUTION]: The skill relies on dynamic execution as a primary capability, using interceptor eval for browser-based JavaScript execution and interceptor macos intent dispatch for AppleScript execution on the host system. It also leverages subprocess APIs in Tools/FrameScrub.ts and Tools/VerifyImageProbe.ts to execute system binaries like ffmpeg and magick.
[INDIRECT_PROMPT_INJECTION]: The skill ingests raw DOM content, accessibility trees, and text from external web pages. This creates a surface for indirect prompt injection, as there are no explicit boundary markers or automated sanitization routines defined for the ingested data before it is processed by the agent.
[PERSISTENCE]: The skill installs LaunchAgent plists (com.interceptor.bridge.plist and com.interceptor.daemon.plist) into the user's ~/Library/LaunchAgents/ directory to ensure background bridge and daemon services persist across system restarts.
[COMMAND_EXECUTION]: The skill frequently executes shell commands and system utilities, such as launchctl for service management, rsync for file synchronization, and codesign for managing binary signatures.
[EXTERNAL_DOWNLOADS]: The Workflows/Update.md script instructs the agent to pull updates from a project GitHub repository and use bun install to download external dependencies during the build process.
[DATA_EXFILTRATION]: The skill provides tools for capturing and exporting sensitive information, including network traffic headers, browser cookies, system logs via macos log query, and filesystem contents.

do-interceptor