Fix: Proposal: add quote display sanity checks for Base MCP agent

skills/base-mcp/references/approval-mode.md

@@ -35,3 +35,4 @@ In approval mode, every write call (send, swap, sign, batched calls, and any plu
 - Polling the status tool in a tight loop instead of once after the user confirms.
 - Forgetting to also auto-open the link in CLI harnesses where a shell is available — printing alone makes the user copy-paste unnecessarily.
 - Trying to auto-open in chat-only harnesses where no shell exists — that just produces an error.
+- Displaying a swap/DeFi quote or transaction amount summary without verifying the human-readable value against the raw wei amount and the token's decimals — see [quote-sanity.md](quote-sanity.md) before showing any quote or approval summary.

skills/base-mcp/references/quote-sanity.md

@@ -0,0 +1,48 @@
+---
+title: "Quote Display Sanity Checks"
+description: "Skill reference for verifying human-readable token amounts before showing a quote or requesting approval."
+---
+
+# Quote Display Sanity Checks
+
+Before presenting a swap / DeFi quote, or asking the user to approve a transaction, verify that any human-readable token amount shown in the chat summary was actually derived from the raw amount using the token's correct `decimals`. Do not eyeball, mentally shift a decimal point, or reuse a value computed earlier in the conversation without recomputing it.
+
+## The problem this addresses
+
+Even when the underlying calldata and the Base Account approval screen are correct, the agent's chat summary is still part of the user's trust boundary. If the displayed human-readable amount doesn't match the raw amount, a user reading the summary before opening the approval link may misunderstand what they are about to approve.
+
+## Format function
+
+Use this conversion whenever a raw token amount (in the token's smallest unit, e.g. wei) needs to be shown to the user:
+
+```
+function formatTokenAmount(rawAmountWei, decimals):
+    # rawAmountWei: integer amount in the token's smallest unit (e.g. wei for ETH)
+    # decimals: the token's decimals value (18 for ETH, 6 for USDC, etc.)
+    return rawAmountWei / (10 ** decimals)
+```
+
+`decimals` must come from token metadata or the MCP response for that specific token — never assume it, and never assume it matches another token's decimals in the same quote (e.g. USDC has 6 decimals, ETH has 18).
+
+## Worked example (matches the reported issue)
+
+- Raw output amount: `50355726366371` wei
+- Token: ETH, decimals: `18`
+- Human-readable: `50355726366371 / 10^18 = 0.000050355726366371 ETH`
+
+An agent summary that instead displays `0.050356 ETH` is off by roughly three orders of magnitude and must not be shown to the user.
+
+## Checklist before showing a quote or approval summary
+
+1. Identify the raw amount and the token's `decimals` for every amount you're about to display (input and output).
+2. Apply `formatTokenAmount(rawAmountWei, decimals)` explicitly — do not shortcut this.
+3. Sanity-check by reversing the calculation: `humanReadableValue * 10^decimals` should equal the original raw amount.
+4. Only after this check passes, present the quote or approval request summary to the user.
+5. If you cannot confirm the token's `decimals`, say so and ask, or fetch it — do not guess.
+
+## Common mistakes
+
+- Off-by-one-order-of-magnitude errors from misremembering or guessing a token's `decimals`.
+- Manually reformatting a raw wei string (e.g. inserting a decimal point by eye) instead of doing the actual division.
+- Reusing a partially-computed or earlier-in-conversation value instead of recomputing from the raw amount for the final summary.
+- Applying one token's `decimals` to a different token's raw amount in a multi-token quote (e.g. using USDC's 6 decimals when formatting the ETH output amount).