Starter prompts for coding agents

Paste one of these into your coding agent with this repo cloned or its llms.txt loaded.

1. Research before building

Use the Ed's Agentic Finance Rails wiki (start at llms.txt and index.md). For <broker>,
tell me: endpoint, auth, whether an agent can place orders, the approval model, and
paper-trading support. Cite the wiki page and the source card for every fact. If the
wiki says "unknown" or has no source, say "unknown". Do not guess.

2. Pick a broker for a use case

Read data/providers.json and approval-models/. I want an agent that <use case>. Rank the
providers by how much of the approval is enforced by the broker rather than my client.
Quote each provider's approval_model and can_place_orders values exactly and cite the
provider page. Flag every "unknown" field that matters for my use case.

3. Build a client with a human approval step

Using build/human-in-the-loop.md and the provider page for <broker>, write a minimal
MCP client that connects to the documented endpoint, lists tools, and never sends an
order without showing me the full order summary and getting my explicit yes. Start on
the paper or sandbox endpoint if the provider page lists one. Cite the pages you used.

4. Check the wiki is current

Before relying on <broker>'s page, check its last_verified date and page_status. If
page_status is "stale" or "archived", or last_verified is older than 30 days, warn me
and open the source cards it cites to confirm the key facts.

If you find something wrong or missing, use the “Suggest a correction” link at the bottom of any page.

---

Suggest a correction (opens a GitHub issue)