The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's fallback protocol instructs the agent to dynamically download and install code from external, non-verified sources using shell commands at runtime.
The instructions in the Auto-Recovery Protocol section explicitly state: "The agent autonomously reads fallback_references, installs/invokes the fallback skill on-demand via npx skills add ."
[COMMAND_EXECUTION]: The instructions require the agent to execute shell commands autonomously to handle task failures, specifically npx skills add web-infra-dev/midscene-skills@computer-automation and npx skills add am-will/codex-skills@gemini-computer-use.
[EXTERNAL_DOWNLOADS]: The skill triggers network requests to fetch and install software packages from untrusted third-party repositories via the npx commands defined in fallback_references.
[PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by processing data and elements from arbitrary web applications during its primary operation.
Ingestion points: Interacting with web apps, clicking buttons, filling forms, navigating pages, and capturing screenshots (SKILL.md).
Boundary markers: Absent; no delimiters or warnings to ignore instructions embedded in the target web content are provided.
Capability inventory: Command execution via npx, browser navigation, form interaction, and UI manipulation.
Sanitization: Absent; no escaping or filtering of external web content is defined before the agent processes it.

computer-use