The Agent Skills Directory

[DYNAMIC_EXECUTION]: The helper script scripts/lib/browser.mjs dynamically resolves and imports the playwright or playwright-core library using computed paths derived from the current working directory and module location. This allows the skill to load and execute code from the local environment, which could result in the execution of malicious code if a package with a matching name is present in the project being inspected.
[INDIRECT_PROMPT_INJECTION]: The skill fetches and processes content from external websites via the inspect-metadata.mjs script, creating a surface for indirect prompt injection attacks.
Ingestion points: External URLs processed by loadUrlSite in scripts/lib/pages.mjs and metadata extracted in scripts/inspect-metadata.mjs.
Boundary markers: Absent. The skill output does not use delimiters or explicit instruction-isolation techniques when passing external content to the agent.
Capability inventory: Network requests via fetch, headless browser automation via Playwright, and local file reads (e.g., package.json, vercel.json).
Sanitization: The skill parses HTML structure using a custom parser but does not apply safety-focused sanitization to the extracted text before it is returned to the agent context.
[COMMAND_EXECUTION]: The skill invokes local Node.js helper scripts and manages headless browser instances. These capabilities provide a significant execution surface for the agent during the reconnaissance process.

site-reconnaissance