Skip to main content
You’ve connected Claude and your permission grants are on the defaults. You want to bump your Mythic rule from a 1.10x multiplier to 1.15x. This is the walkthrough for what that actually feels like the first time you do it. The flow has three beats: you ask, Claude shows you the preview, you click to approve. Total time, maybe 30 seconds.

​
Before you start

  • A paid plan (Gold, Platinum, or Diamond), Claude connected via the OAuth connector
  • At least one pricing rule that targets Mythic rarity (or substitute whichever rule you actually want to edit)
  • Permission grants left on defaults. The relevant one here is pricing.edit_rule: ask_for_mutations. If you tightened that to always_ask, the flow is the same; you just get one prompt regardless of risk score.

​
Step 1: Ask Claude

Open Claude.ai with the Hoard connector enabled. Type:
“Bump my Mythic rule from 1.10x to 1.15x.”
Claude prepares a proposal behind the scenes. It does not change the rule yet. The server derives how many cards would shift, the projected aggregate change in listed value, the risk classification, and a Hoard review URL. Claude shows the price change preview inline: the risk pill, the dollar impact, and the confirmation link. What you should see in the response:
  • Cards affected. Something like “180 cards.” This is the count of inventory rows the rule currently selects.
  • Projected change. Something like ”+$72.40 listed value.” Positive because you’re raising the multiplier. If it were negative and big enough, a hard floor might trip, see When the agent says no.
  • Risk pill. Probably “low risk.” A 1.10 → 1.15 change is a +4.5% multiplier move, well under most sellers’ change_cap_pct.
  • Confirmation link. A URL like https://www.tryhoard.com/agent/plans/<plan_id>.
If you ever see “high risk” or “critical risk” on a small-looking edit, stop and read what Claude wrote. Risk scoring is user-relative, see Picking the right permission mode for what tightens it. Click the link Claude gave you. It opens in whichever browser you’re already signed in to Hoard with. No second login. You land on a plan detail page that shows the same information Claude saw, plus:
  • The recognizable rule name and query
  • The exact old and new multiplier, price source, never-go-down setting, and enabled state
  • The number of cards affected and projected listed-value impact
  • Protected-card warnings and the overall risk rating
  • An Approve and apply button and a Reject button with optional feedback
If Claude proposed multiple related rules, they appear together on this page. You approve or reject the whole matrix; there is no partial approval. Read the diff. If anything looks off, wrong rule, wrong direction, way more cards than you expected, click Reject. The plan dies, Claude gets told, nothing committed. If it looks right, click Approve.

​
Step 3: Hoard applies the proposal

Clicking Approve and apply performs the write immediately on Hoard’s servers. Claude does not need to stay connected, receive a browser token, or retry the request. When the page shows Applied, the change is done. The whole proposal is atomic. If a rule was edited after Claude prepared the proposal, or any item now fails a safety check, none of the changes apply. The page explains that the proposal is stale or failed; return to the chat and ask Claude to refresh it. Back in your Claude chat, you’ll see something like:
“Done. Updated your Mythic rule from 1.10x to 1.15x. 180 cards repriced. Your next sync will push these to TCGplayer.”
The change is now live in Hoard. Your next sync cycle pushes it to TCGplayer. Claude can read the proposal’s durable status and report the applied change, the card count, and a note that your next sync pushes it to TCGplayer.

​
Trying a multi-rule request

Once the single change feels familiar, try:
“Review my current pricing rules and propose a cleaner strategy for bulk, foils, and Reserved List cards. Don’t change anything until I approve it.”
Claude can return one proposal containing new rules, edits to existing rules, and enable/disable changes. The review page groups them into one readable matrix. One click applies the exact matrix together, and one rejection applies nothing. Proposals expire after one hour, so ask Claude to regenerate an old link rather than approving stale assumptions. You do not need to recreate the proposal or hunt through Settings. Ask your assistant:
“Find my pending pricing proposals.”
The assistant can list your pricing change sets, recover the original review link, and summarize every rule in the matrix. It can narrow the search to pending, applied, rejected, stale, failed, or expired proposals. If it retained the proposal fingerprint, it can use that to find the exact proposal even when several were created close together. Retrying the original request with its same idempotency key also returns the same proposal instead of creating a duplicate. The key remains reserved while Hoard retains the proposal record—currently seven days after expiry. After that retention window, ask the assistant to inspect current rule state and create a fresh proposal with a new key. An expired, rejected, or stale proposal never silently reopens. The assistant reports its durable status and, when appropriate, prepares a new proposal from current store data for you to review.

​
Where the change shows up later

Two places. Your dashboard. The rule now reads 1.15x where it used to read 1.10x. Repricing has either already happened (if you’re on autonomous sync) or is queued for your next confirm-mode sync. Your agent activity log. This is the audit trail. The easiest way to find this specific change later is to ask Claude: “what did you change for me today?” — it pulls from the activity feed at GET /api/v1/agent/activity and summarizes. Filter by namespace=pricing.edit_rule&outcome=committed&since=<date> if you’re hitting the endpoint directly. The row tells you the before/after multiplier in params_summary, what the risk score was, and links back to the plan detail page via agent_action_plan_id if you want to re-read it. A dedicated Activity tab in Settings → Assistants is on the roadmap. For a full tour of the activity log including every outcome type, see Reading your agent activity log.

​
What if you’d loosened the grant?

If you’d set pricing.edit_rule to auto_low_risk, this exact edit would have skipped steps 2 and 3 entirely. Claude would have called preview, seen the low-risk classification, called commit on the same turn, and reported back “done” in one round trip. No confirmation link, no click-through. You’d still see it in the activity log. Floors still gate. Risk scoring still runs, if the same conversation had asked for a 5.0x multiplier, the call would have escalated to confirmation regardless of the loosened grant. That’s the trust progression: start with confirmations, watch the patterns, loosen the dials once you trust them. See Going from cautious to confident.