The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses npx -y and uvx to fetch and execute tools from public registries (npm and PyPI). Specifically, it references chrome-devtools-mcp@latest, windows-mcp, and devstudio-mcp. The use of @latest in the npx command avoids version pinning, which can lead to the execution of unverified updates.
[COMMAND_EXECUTION]: The instructions mandate 'Zero-Bailout Auto-Provisioning,' which explicitly directs the agent to execute shell commands (npx, uv tool run, uvx) to install missing dependencies without manual user intervention.
[REMOTE_CODE_EXECUTION]: By automatically triggering the download and execution of external packages at runtime, the skill performs remote code execution to enable its core functionality.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection as it is designed to ingest and process live data from external web applications and desktop environments.
Ingestion points: The agent navigates to web URLs via navigate_page and interacts with desktop windows via list_windows and focus_window.
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded commands within the ingested application content.
Capability inventory: The skill possesses capabilities to write files to the local file system (docs/manual/), take screenshots, and record screen/audio content.
Sanitization: The skill includes a 'Quality Review' step to scrub credentials and private data from outputs, but lacks specific sanitization for instructions embedded in the target application's text or metadata.

app-tutorial-manual-generator