Proposal: add quote display sanity checks for Base MCP agent approval flows

Hey dear Base team 💙

I’d like to propose a small safety improvement for the Base MCP skill: adding explicit quote display sanity checks before an agent presents a swap / DeFi quote or asks the user to approve a transaction.

While testing Base MCP through Hermes + Telegram, I noticed a quote display issue during a Uniswap swap flow.

The agent was asked to prepare a quote for:

0.1 USDC -> ETH on Base

In the Telegram chat, the agent displayed the raw output amount in wei correctly, but the human-readable ETH amount appeared to be formatted incorrectly.

Example:

* Raw output: 50,355,726,366,371 wei
* Correct human-readable value: 0.000050355726366371 ETH
* Displayed in the agent summary as: 0.050356 ETH
Important clarification: the Base Account approval screen appeared to show the transaction amount correctly. So this does not seem to be a Base Account execution issue, and it may not be a Base MCP backend issue either.
The issue was observed in the Hermes + Telegram agent-facing summary before approval. I did not reproduce the same behavior in ChatGPT.

Why this matters
Even if the underlying calldata and Base Account approval flow are correct, the chat summary is still part of the user’s trust boundary.
A user may read the agent’s summary before opening the approval link, and if the displayed human-readable token amount is wrong, they may misunderstand what they are about to approve.
This seems especially relevant for multi-surface MCP usage, where Base MCP can be used through different clients such as Hermes, Telegram, Claude, ChatGPT, custom MCP clients, and other agent environments.

Proposal
Add a new reference file, for example:
skills/base-mcp/references/quote-sanity.md
The goal would be to instruct agents to validate raw token amounts against human-readable values before showing a quote or requesting approval.

Start by reviewing the Base MCP skill structure and the proposed skills/base-mcp/references/quote-sanity.md path. Use the raw and human-readable amount example in this issue to define the sanity-check guidance. Done means the reference clearly instructs agents to validate token amounts before displaying quotes or requesting approval.

Written by the indexing model from the issue text.