forum-research

Forum & community research (Reddit, Hacker News, project forums)

When to Use

• "Search Reddit and other forums carefully", "is there a simpler alternative to X", "is X worth using", "does X actually work in practice" — the community verdict and real-world experience, not the project's own docs.

Procedure

1. Search the pain, not the product. Query the constraint that made the user look ("simpler than virt-manager", "no root", "instead of a VM", "no libvirt", "throwaway VM") — the product name alone returns release announcements and one-line praise. Run the name query plus 2-3 pain queries.
2. Reach the sources through their APIs, not their HTML (exact URLs + a copy-paste fetcher: references/community-apis.md):
   • Reddit → the PullPush API (submission search, comment search, and link_id=t3_<id> for one thread's comments).
   • Hacker News → the Algolia API (tags=comment|story, then /api/v1/items/<id> for the whole thread).
   • Project/vendor forums, AUR comment threads, GitHub issues and wikis → ordinary web_search + web_extract; this is where version-specific breakage and packaging caveats live (the ones worth reading directly are listed at the end of references/community-apis.md).
3. Fetch with curl/python3 in the terminal, one query per request, each response written to a JSON file under the scratch dir; aggregate and filter in a second pass (curl, python3, jq are all present).
   • Don't loop fetches inside one browser_exec js() await — the CDP timeout (~5 s) kills it mid-loop; batch work belongs in the terminal.
   • Keep the query list as (name, kind, query) tuples and call the fetcher inside the loop. Splatting a dict of results back into the fetcher silently feeds a 25-element list as positional args and wastes the run.
4. Filter before printing so context isn't flooded: top items by score as subreddit + score + ≤300-char body, dropping bodies under ~60 chars (link-only noise). Keep the raw JSON on disk so a follow-up question re-reads it instead of re-fetching.
5. Report short. Lead with the answer ("yes — X, and it removes A and B"), then 2-4 ranked bullets: option → what it removes → one community quote as evidence → the tradeoff. The long tail (full quotes, links, command recipes) goes into the repo doc or skill, never the chat — this user reads a few sentences per point, not paragraphs.