The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes and analyzes HTML content from external websites to audit form functionality, validation, and accessibility. This behavior introduces a surface for indirect prompt injection where instructions hidden in the analyzed site's markup could attempt to influence the agent's behavior or reporting.
Ingestion points: The scripts audit-markup.mjs and check-links.mjs fetch and parse HTML from user-specified URLs or local build directories.
Boundary markers: The skill does not explicitly use protective delimiters or specific warnings to the agent to ignore instructions embedded within the audited data.
Capability inventory: The skill has the ability to make network requests, write screenshots to the disk, and drive a headless browser.
Sanitization: Data is parsed into an abstract tree for analysis, but the resulting text output is passed back to the agent without specific sanitization against instruction-like patterns.
[DYNAMIC_EXECUTION]: The skill uses dynamic loading techniques to locate and initialize browser automation libraries at runtime based on the local environment.
Evidence: scripts/lib/browser.mjs programmatically resolves the path to playwright or playwright-core and uses import() to load the dependency, which is a form of dynamic loading from computed paths.
Script interpretation: The observe-runtime.mjs script accepts a JSON file defining a sequence of browser actions (e.g., click, fill, submit), which it interprets and executes at runtime.

forms-readiness