Smarter back button + board context for AI agents

Two small improvements that go together — one polish, one quietly leveraged feature for the AI parts of Simple Product.

Back button now goes back to where you came from

Previously, the back arrow on a person's page or an organization's page always took you to the corresponding listing — even if you got there by clicking through from another record. So if you went Person → linked Org → back, you ended up on the Orgs listing instead of returning to the Person you were on. Surprising, mildly disorienting, and a quiet papercut every time you navigated through linked records.

Now the back button checks whether you got here from inside Simple Product. If yes, it returns you to wherever you came from. If you landed via an external link (Slack, email, refreshed the page), it falls back to the listing — same as before. So:

• Person → Org → Back returns to the Person ✓
• Org → Person → Back returns to the Org ✓
• Listing → Detail → Back returns to the Listing ✓
• Email/Slack link → Detail → Back returns to the Listing (no in-app history) ✓

Same fix should land on other detail pages (cards, docs, releases) — that's a quick follow-up.

Boards can now hold context for AI agents

Each board can now carry an optional context — a small free-form description of what the board is for, who works in it, or how cards on it are formatted. Examples:

This board tracks customer outreach, not features. Cards follow Name → Company → Status → Last touch.

This is for raw ideas to research; don't expect them to be well-formed.

Engineering bug board. Cards link a reproducible repro and a stack trace.

The context is hidden in the UI by design — there's no header banner showing it on the board page. Instead, it's stored alongside the board and surfaces in agent prompts when AI is working with that board's cards. The agent reads the context, adapts its behavior, and produces output that fits the conventions of the specific board.

That last part is the load-bearing piece. Today, when you ask Claude or any MCP-connected agent to "draft a card on the customer outreach board," it has no way to know that board has different conventions than your features board. Context closes that gap without forcing every prompt to re-explain it.

How to use it

Setting context when creating a new board: the create dialog now has an optional Context textarea below the name. Skip it and you get the same lightweight board you've always created. Fill it in and AI agents working on the board pick up the conventions.

Setting context on an existing board: open the board, hit the [⋮] menu in the header, click Add context (or Edit context if one's already set). Same textarea, same behavior. Save it and you're done.

Clearing context: open the same dialog, empty the field, save. The menu label flips back to "Add context."

Note on the AI side

The data and UI for context are live in this release. The wiring that injects board.context into agent prompts is queued as a follow-up — small change, planned soon. Any context you write today is captured and waiting for that wiring to land. No need to redo it later.

The MCP get_board tool will also include context in its response after the next MCP server-side update — happens automatically; no SDK reinstall needed (this is the "ship without npm publish" architecture from the V1 MCP rebuild).