yichengyang-ethan/oracle3/pm-data-discovery

PM Data Discovery

Use this skill when the user wants to look at data, find markets, or collect research samples before any strategy work.

Goals

Answer three questions:

• Which data sources are available?
• How do I find the target markets?
• How do I save the data as files that later strategy research can use?

Data entry points

1. Market metadata and search (online)

• oracle3 market list --exchange polymarket --limit 50 --json
• oracle3 market search --exchange polymarket --query "<keyword>" --limit 50 --json
• oracle3 market info --exchange polymarket --market-id <market_id> --json
• oracle3 market history --market-id <market_id> --interval 1h --limit 500 --json

1. News and external events (online)

• oracle3 news fetch --source google --query "<keyword>" --limit 30 --json
• oracle3 news fetch --source rss --query "<keyword>" --limit 30 --json

1. Local history files (offline)

• oracle3 research markets --history-file <history.jsonl> --sort-by points --limit 50 --json
• oracle3 research slice --history-file <history.jsonl> --market-id <M> --event-id <E> --output <slice.jsonl> --json

1. Raw stream recording (online)

• oracle3 data record --exchange polymarket --output <events.jsonl> --duration 3600 --json

1. Through the MCP server (oracle3-mcp)

• search_markets, get_market, get_orderbook, get_quote for Kalshi and Polymarket.

Recommended flow

1. Use market search or market list to define the candidate markets.
2. Pull market info and market history for the candidates to check liquidity and price behavior.
3. For offline research, record with data record or use an existing history file.
4. Produce slices of the key markets for the strategy agent to read.

Output requirements

• A clear data inventory: markets, time range, file paths.
• Every conclusion comes with a command that reproduces it.
• If the network is unavailable, switch to the local research commands instead of stopping.

Hard rules

• Do not propose strategy logic in this skill.
• Only discover, filter, sample and save data.
• Prefer --json output so that later agents can consume it.