Perplexity research

Skill zacharygcook/agent-skills/skills/perplexity-research

A curated collection of practical, evidence-backed skills for coding agents.

Install

npx -y skills add zacharygcook/agent-skills --skill perplexity-research

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

26 days oldThe repository was created 26 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use Perplexity MCP for current, cited web research. Trigger when the user asks to use Perplexity, asks for deep research, market/vendor/provider research, current comparisons, current API/service capability checks, or wants citation-backed investigation beyond ordinary browsing.

SKILL.md

3.1 KB, as published. Nobody here has run it

Perplexity Research

Use the Perplexity MCP server for research that benefits from current web search, source synthesis, and citations. Prefer it for vendor/provider comparisons, service capability research, recent API documentation checks, and broad market scans.

MCP Setup Assumption

The local Codex MCP server is named perplexity.

If Perplexity tools are not already exposed, discover them with tool_search using a query like perplexity research. Expected tools from the MCP server:

perplexity_search: ranked web search results.
perplexity_ask: quick web-grounded answers.
perplexity_research: deeper cited research reports.
perplexity_reason: reasoning-heavy analysis.

If the server fails because PERPLEXITY_API_KEY is missing, tell the user to add a real key to ~/.perplexity-mcp.env and restart Codex/MCP. Do not invent Perplexity results if the MCP is unavailable.

Tool Choice

Use perplexity_research for comprehensive vendor/provider research, due diligence, competitive comparisons, or questions with high business impact.
- Deep research can take several minutes. If the current tool surface times out around 300 seconds and does not expose a per-call timeout, treat that as an environment/tooling limit, not a research answer.
- On deep-research timeout, fall back to perplexity_reason with search_context_size: "high" for comparative analysis, or perplexity_ask with search_context_size: "high" for source-grounded factual summaries.
Use perplexity_search first when you need a short source list before deeper investigation.
Use perplexity_ask for narrow factual questions.
Use perplexity_reason when the user asks for strategy, tradeoffs, ranking logic, or decision support after evidence has been gathered.

Research Standards

Prefer primary sources: official docs, pricing pages, API references, provider terms, and public product pages.
Use secondary sources only to fill market context, not as proof of API capability.
Distinguish confirmed facts from inference.
Preserve source URLs in the final answer.
For API/vendor work, verify:
- supported input identifiers,
- response objects/fields,
- whether the provider finds people, relatives, owners, heirs, or only enriches known people,
- pricing model,
- API availability,
- batch versus synchronous mode,
- fit for the user's actual workflow.

Output Shape

For vendor/provider research, return:

Short conclusion.
Ranked options with best use case and caveats.
Table of confirmed capabilities.
Recommended workflow or waterfall.
Open questions to verify with sales/support or API samples.
Source links.

Do not overstate "heir" capability. If a provider returns relatives, household members, associates, owners, or enriched person profiles, label that precisely.