The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and process data from external, untrusted web pages, which serves as a significant attack surface for indirect prompt injection.\n
Ingestion points: Data is retrieved through WebFetch, Browser automation tools, and Bash-based curl requests as defined in SKILL.md and references/extraction-patterns.md.\n
Boundary markers: The SKILL.md file contains a dedicated 'Security' section that explicitly instructs the agent to treat all scraped content as data, ignore instructions embedded in that data, and require user confirmation for navigation or code execution suggested by the content.\n
Capability inventory: The skill allows the agent to execute shell commands (Bash), run JavaScript within a browser context, and perform file system writes for data export.\n
Sanitization: The skill provides instructions for cleaning data, including the removal of zero-width characters and Unicode normalization in references/data-transforms.md.\n- [DYNAMIC_EXECUTION]: The skill involves the execution of dynamically generated or selected JavaScript snippets in the browser and Python one-liners via Bash to process extracted data, as seen in references/extraction-patterns.md and SKILL.md.\n- [COMMAND_EXECUTION]: The skill uses shell commands like curl and jq to interact with web APIs and process JSON data.

web-scraper