When a machine must ingest a public CSV URL into structured rows for an automated load or transform step, Task Relay fetches UTF-8 CSV and returns ordered row arrays with column and dialect metadata under fixed parse limits. No type inference, Excel semantics, database loading, or AI enrichment.

When a machine already holds an HTML string and needs deterministic Markdown for a content pipeline, Task Relay converts the HTML under fixed server rules and returns Markdown plus source hash, structural counts, output hash, truncation, and warnings. No URL fetch, no browser rendering, no JavaScript execution, no model inference or rewriting.

When a machine needs a fixed visual record of a public HTTP or HTTPS page for automated archive or rendered-page indexing, Task Relay renders the URL in a bounded managed browser runtime and returns one fixed-viewport JPEG plus navigation and artifact metadata. No client browser options, selectors, scripts, cookies, login, or full-page capture.

When a machine needs plain text from a public PDF URL for ingestion or indexing, Task Relay fetches the PDF and extracts text-layer content page by page into a statused result with text, page counts, hashes, truncation flags, and warnings. Text-layer extraction only; NO OCR; no summarization; scanned image-only PDFs may return status no_text.

When a machine must decide whether to continue, defer, or reject work on one public HTTP or HTTPS URL, Task Relay performs a single bounded reachability observation (DNS, redirects, HTTP status, TLS validity, sampled content fingerprint, elapsed time) and returns a strict statused machine-readable result for automated gating. Single-shot observation only; not continuous monitoring and not page extraction.