Web Search Playbook

Use this skill whenever the user needs up-to-date information, troubleshooting steps sourced from the web, or comparisons that aren’t already in the repo. It orchestrates whatever search tools your repo provides (e.g., browser-tools for Chrome automation, lynx for text-mode browsing) so you can search efficiently and report trustworthy results.

When to Use

• •User explicitly asks you to “look it up,” “check Hacker News,” “see what people recommend,” etc.
• •You need confirmation of facts, recent news, pricing, or documentation outside the local project.
• •Troubleshooting production errors where error codes or stack traces should be searched verbatim.

Skip this skill if the answer is clearly available in the current repository/docs or if the user forbids external browsing.

Tool Selection

Always mention which tool was used and save captures (e.g., /tmp/search-result.txt) when results will be referenced later.

Workflow

1. •

   Clarify Objective

   • •Restate the user’s question and ask for missing parameters (time range, geography, stack version) before searching.
   • •Example prompt: “To narrow results, do you care about the last 6 months or any time?”
2. •

   Craft Strong Queries

   • •Start specific: include error codes, API names, and key symptoms.
   • •Use quotes for exact phrases; use -term to exclude noise.
   • •Apply operators when helpful: site:docs.oracle.com, filetype:pdf, intitle:"error 500".
   • •Try synonyms (e.g., “pagination cursor” vs. “infinite scroll offset”).
3. •

   Execute Searches

   • •DuckDuckGo Lite via lynx:
     bash

     lynx -dump -nolist "https://duckduckgo.com/lite/?q=<encoded query>"
     

   • •Browser-based search (if a visual SERP or login is required):
     bash

     ./browser-nav.js "https://www.google.com/search?q=<query>"
     

   • •Capture the SERP output (text dump or screenshot) for reference.
4. •

   Triage Results

   • •Prioritize official docs, reputable blogs, recent timestamps, and community answers with clear acceptance.
   • •Open the top 2–4 promising links; skim headings before deep reading.
   • •If all matches are low-quality, reformulate the query.
5. •

   Extract and Record Evidence

   • •For text-friendly pages:
     bash

     lynx -dump -nolist <url> > /tmp/<slug>.txt
     

   • •For dynamic pages:
     bash

     ./browser-content.js <url> > /tmp/<slug>.md
     

   • •Note publication dates, author credibility, and key quotes in your notes (or findings.md for long tasks).
6. •

   Ask the User When Needed

   • •If the search surfaces multiple conflicting approaches (“Node vs. Python solution”), summarize trade-offs and ask the user which direction to pursue.
   • •When results hinge on environment details (cloud provider, OS), confirm before acting.
7. •

   Iterate

   • •Use new keywords found in articles (library names, error IDs) to refine the query.
   • •Switch engines (e.g., GitHub search, Stack Overflow, vendor docs) if the general web is noisy.
8. •

   Summarize with Citations

   • •Provide at least two independent sources when possible.
   • •Include direct URLs and the date accessed.
   • •Call out uncertainties or remaining questions and suggest next searches if needed.

Practical Tips

• •Time filters: add past year or past month options on DuckDuckGo/Google if results might be stale.
• •Language filters: append the framework/language ("react 18" suspense) to avoid irrelevant hits.
• •Notebooking: for multi-hour research, keep a running bullet list of URLs + takeaways in findings.md so you don’t revisit the same pages.
• •Conflicting sources: explicitly state disagreements and why you prefer one (e.g., official docs dated 2026 vs. an outdated 2019 blog).

Hand-off Checklist

• •State what was searched (queries, operators, time filters).
• •Enumerate sources with short annotations.
• •Attach/cite any saved dumps or screenshots from lynx/browser-tools.
• •Flag any assumptions you made so the user can correct them.

Learned Lessons

Add new heuristics here whenever you discover better querying patterns, niche documentation sources, or ways to combine search tools like browser-tools and lynx more effectively.