The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of a global npm package named dev-browser from the public npm registry. This package is not pinned to a specific version, and the skill documentation notes that the tool may download an additional binary during its first run if the initial installation was restricted.
[COMMAND_EXECUTION]: The instructions involve running shell commands to install the tool and execute automation scripts, which are necessary for the skill's stated purpose but involve the execution of external binaries.
[DYNAMIC_EXECUTION]: The CLI tool executes dynamic Puppeteer scripts (JavaScript) provided via standard input or command-line arguments. This allows the agent to generate and execute code at runtime to interact with the browser.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to browse the web and extract data (scraping, ARIA tree snapshots, screenshots). This creates a vulnerability where the agent might ingest and act upon malicious instructions hidden within the web content it processes.
Ingestion points: External web content retrieved by the browser automation scripts (SKILL.md).
Boundary markers: None present; the instructions do not define delimiters or warnings to ignore embedded instructions in the scraped data.
Capability inventory: The skill includes network access for browsing, file system writes for saving screenshots/PDFs, and file system reads for uploading files (SKILL.md).
Sanitization: No sanitization or validation logic is specified for the data extracted from the browser.

dev-browser