The Agent Skills Directory

[PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection as it processes untrusted data from external websites.
Ingestion points: HTML, text, and element metadata retrieved from target URLs via agent-browser as described in SKILL.md.
Boundary markers: No explicit delimiters or instructions are used to separate scraped content from agent instructions.
Capability inventory: The skill uses the Bash tool and browser automation capabilities.
Sanitization: No evidence of sanitization or filtering of the scraped content is present.
[COMMAND_EXECUTION]: The skill utilizes shell commands through the Bash tool to create local directories for the website blueprint and to download assets.
[EXTERNAL_DOWNLOADS]: The skill is configured to download various file types (images, videos, fonts) from user-specified target websites during the documentation phase.
[SAFE]: The skill includes positive security features such as a mandatory user confirmation step for legal authorization and explicit instructions to avoid scraping behind authentication walls or collecting personal data.

glmv-web-replication