The AI working your board
Short answer: The AI reads, adds, edits, expands, and deletes cards through five card tools, and every add, edit, or delete shows up as an approval banner in the Lists panel before anything changes. You can also paste content and ask it to turn that into cards, or send a whole list into chat as context.
What tools does the AI use on my board?
Grouped under Cards in Build Perch’s tool catalog:
add_card: propose one or more new cards (title and notes, plus optional category, priority, deadline, stage, and URLs).read_cards: read a list’s cards, grouped by stage; the AI’s default way to see what’s on the board.expand_card: pull full detail for specific cards: notes, URLs, attached files, and a per-stage time breakdown.edit_card: change one or more cards’ fields.delete_card: remove one or more cards, and only when you’ve explicitly asked for it. It only ever removes the card from the list; it never touches files on disk, including a card’s attached files.
A sixth tool, search_cards, does meaning-based (semantic) search over whichever list is currently open: it can match “the deployment card” to a card titled “Production rollout” even without a shared word. It’s scoped to the active list only; switching lists resets it.
Edits and deletes are batchable: one edit_card or delete_card call can cover several cards at once, so “move these five cards to Done” resolves in a single approval round instead of five.
How does approval work for card changes?
Every add_card, edit_card, and delete_card call surfaces as a banner inside the Lists panel; nothing is written until you act on it. While one is pending, the chat bubble tells you a card change is awaiting approval and points you at the Lists panel.
- Adding cards: each proposed card gets its own row with a checkbox, checked by default; uncheck any you don’t want before approving.
- Editing cards: each edit shows a field-by-field diff (the old value struck through, the new value beside it) so you see exactly what would change.
- Deleting cards: each card slated for deletion gets its own checkbox in a red-tinted block; uncheck any you want to keep.
Only one banner is on screen at a time. Edits are also scoped to the fields you named, and the field-by-field diff in the approval banner shows exactly what would change before you approve, so an edit request like “bump the priority to 2” won’t quietly blank out a card’s title or notes.
Can the AI turn pasted content into cards?
Yes. Paste something into chat (a meeting transcript, a bug list, a rough plan) and ask the AI to turn it into cards. It reads the text and proposes one card per item via add_card, all in a single batch you approve at once. New cards land in whichever list you have open in the Lists panel, so switch to the right list first; you can still tell the AI which stage each card should start in.
Turning on the Lists context skill (in the Mode pill’s skill list, under Context Injectors) has the AI start every turn already knowing your lists’ names, stages, and card counts, which saves it a read_cards round-trip before it can act.
How do I send a list into chat as context?
Click Send List to Chat at the bottom of the Lists panel. It attaches a formatted snapshot of the current list (cards grouped by stage, with priority, category, flag, deadline, notes, reference URLs, and attachment names) to your next chat message. The snapshot is taken when you click; if the list changes afterward, the attachment doesn’t update.
Good to know
- The card tools only appear once the active workspace has at least one list.
search_cardsfinds cards by meaning, even when you don’t remember the precise wording.expand_cardcan surface how long a card has sat in its current stage, alongside its full notes, URLs, and attachments; useful for “what’s stuck” questions.