igorwarzocha__opencode-workflows/scripts/.opencode/command/perplexica-search.md

description

description
Run Perplexica searches with smart mode selection and structured citations

Arguments

• $ARGUMENTS contains the full natural-language query (context + desired outcome).
• If $ARGUMENTS is empty or unclear, ask the user to restate the question.

CRITICAL – follow these steps exactly

Steps

1. Parse the user request and restate the research goal (topic + desired output).
2. Choose the best Perplexica focus mode:
   • webSearch – general news or mixed sources
   • academicSearch – research papers, reports, data
   • writingAssistant – drafting or outlining content
   • redditSearch – community sentiment or discussions
   • youtubeSearch – video explainers or walkthroughs
   • wolframAlphaSearch – factual calculations or reference data
3. Execute the query exclusively through the CLI:

   node Scripts/perplexica-cli.js "$QUERY" --mode $FOCUS_MODE
   

   • The CLI builds the payload, enforces a 300s timeout, and returns JSON with message (summary) and sources (title + URL).
4. Parse the CLI output:
   • Extract the primary answer from message.
   • Collect each source title and URL from sources.
   • Capture the reported runtime (printed by the CLI).
5. Produce the final response:
   • 2–3 sentence summary covering the main findings.
   • Bullet list of the top sources with titles + URLs.
   • Mention number of sources, focus mode, and elapsed time.

Query optimization tips

• Add “latest” or “recent” for up-to-date info.
• Use “how to” for procedural queries.
• Add “comparison”/“vs” for trade-off analysis.
• Use “analysis” or “report” for deep dives.

Sample queries

• “What are the latest NIST controls for AI supply-chain security audits?”
• “How do open-weight LLMs reduce compliance inference costs in 2025 enterprise workloads?”
• “Which newly approved NASA missions will monitor solar storms over the next decade?”

Error handling

• If a specialized mode fails, retry with webSearch.
• Confirm the Perplexica endpoint (default http://localhost:3000/api/search).
• The CLI already retries on transient failures; only rerun if it exits with an error.

Output: Summary paragraph followed by a numbered source list and search metadata.