The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads content from api.microlink.io and v1.screenshot.11ty.dev during the screenshot capture process.
[DATA_EXFILTRATION]: User-provided target URLs are transmitted to external third-party services (Microlink and 11ty) to perform the capture. This exposes potentially sensitive internal or private URLs to these external platforms.
[COMMAND_EXECUTION]: The skill uses multiple shell commands to perform its tasks:
curl is used for network requests and file downloads.
python3 is used to execute a static inline script that parses JSON data from the Microlink API.
magick (ImageMagick) is used to stitch multiple captured frames into a single image.
rm is used to delete temporary image frames from the ~/Downloads/ directory.
[REMOTE_CODE_EXECUTION]: The skill follows a pattern where data returned from the Microlink API is piped directly into a Python interpreter (python3 -c). Although the Python script is hardcoded in the skill, piping network output directly to a language interpreter is a risky pattern that could be exploited if the remote service is compromised or returns malformed data.
[PROMPT_INJECTION]: The skill ingests untrusted data from arbitrary websites via browser navigation and JavaScript execution (mcp__Claude_in_Chrome__javascript_tool).
Ingestion points: mcp__Claude_in_Chrome__navigate and mcp__Claude_in_Chrome__find in SKILL.md.
Boundary markers: None are present to delimit untrusted web content from agent instructions.
Capability inventory: The skill can execute shell commands (curl, python3, magick), write files, and execute JavaScript in the browser context.
Sanitization: No sanitization is performed on the website content before interaction.

capture