Parallel Web Extract

URL Extraction

Extract content from: $ARGUMENTS

Command

Choose a short, descriptive filename based on the URL or content (e.g., vespa-docs, react-hooks-api). Use lowercase with hyphens, no spaces. Substitute it into the command inline — $FILENAME is a placeholder, not a shell variable.

Pass each requested URL as a separate quoted positional argument, up to 20 per call. Do not collapse multiple URLs into one quoted $ARGUMENTS string or use eval to split them. Construct arguments directly from the requested URLs. For example:

parallel-cli extract "https://docs.parallel.ai/integrations/cli" "https://docs.parallel.ai/integrations/cursor-marketplace" --json -o "/tmp/parallel-docs.json"

-o saves JSON. Use a .json extension and inspect an existing path before use because Extract overwrites it. Read the saved file as authoritative; stdout may truncate and human-readable output previews only part of the content. Do not treat a stale file as a successful response after a failed call.

Options if needed:

• --objective "focus area" to focus extraction on a specific goal (also silences the “neither objective nor search_queries” warning that V1 emits when neither is set)
• -q "keyword" (repeatable) to prioritize keywords in excerpts
• --full-content to include the complete page body (for long articles, PDFs, or when excerpts may not capture what you need)
• --full-content-max-chars N to cap full-content size per result
• --no-excerpts to strip excerpts when you only want full content
• --session-id "<returned-session-id>" to group related Search/Extract calls. A session ID is not a Task interaction ID or run ID; never use it with research status/poll or --previous-interaction-id

Handling failed extractions

Inspect the exit status, API error, results, per-URL errors and any warnings. errors: [] is normal success. Nonempty errors can coexist with successful results: retain and present successful content, then name each failed URL and its returned reason. Empty results or missing content are not a successful extraction. Do not fabricate content. For affected URLs, suggest:

• Verifying the URL (the page may have moved)
• Requesting --full-content if excerpts are empty but the returned metadata supports that the page was fetched
• Using parallel-cli search to locate the current URL if the page was renamed

Response format

Return content as:

Page Title (opens in a new tab)

Use returned full_content for full-page requests; excerpts alone are selected passages and must be labelled as such. Even full content may be capped by --full-content-max-chars or upstream limits; do not promise completeness when capped. Preserve retrieved content verbatim, with these rules:

• Keep content verbatim - do not paraphrase or summarize
• Preserve every numbered/bulleted item in the retrieved content; do not claim an excerpt contains the whole page
• Strip only obvious noise: nav menus, footers, ads
• Preserve all facts, names, numbers, dates, quotes

After the response, mention the output file path (/tmp/$FILENAME.json) so the user knows it’s available for follow-up questions.

For large content, keep the full verbatim text in the saved file and provide a brief labelled preview plus its path. Never silently truncate content while claiming it is the complete extraction.

Setup

If parallel-cli or a required option is missing, tell the user which part is missing and offer /parallel:parallel-cli-setup. That skill covers installation and upgrades for each supported install method. After setup, check parallel-cli extract --help in the same terminal before retrying.

For authentication errors, inspect parallel-cli auth --json and its authenticated boolean; exit zero alone does not prove authentication. A 403 can indicate permissions, policy or billing. Report the actual error; check balance only for a billing-specific failure and never add funds without explicit confirmation.