Start with three separate questions
Which application can see an account? Which contract can spend a token? What does the next signature authorize? A wallet interface can present these questions in a similar-looking window, but one answer does not settle the others. This guide is a paper exercise for organizing a review. It does not connect a wallet, inspect an address or certify that an application is safe.
MetaMask distinguishes connected-application permissions from token approvals: disconnecting an application does not remove an allowance already granted to a contract. Its revocation guidance also explains that revoking an onchain approval is itself an onchain action with gas costs. Those two facts explain why closing a browser tab is not a complete permission review.
| Layer | Question to answer | Useful evidence |
|---|---|---|
| Connection | Which account does this application access? | The connected account and application origin. |
| Allowance | Which spender can use which token, and how much? | Token contract, spender, network and current allowance. |
| Signature | What is this particular request authorizing? | Decoded request, destination, amount and stated purpose. |
Make a register before changing anything
Create one row for each permission you intend to review. Use a local label such as “research wallet” rather than putting personal identity into a shared worksheet. Record the network, token contract and spender separately. A familiar token ticker is a display label, while a contract address identifies the particular asset on that network. Keep your private register somewhere appropriate for its contents. Public wallet addresses can still reveal activity.
Write what the interface actually establishes. “No approvals found by this tool” is narrower than “this wallet has no permissions anywhere.” A checker may cover only a particular network, standard or account. Record the network and the tool used so someone else can understand the boundary. An empty cell should say unknown or not checked, not zero.
For each row, describe the intended job in one sentence: for example, “exchange up to 250 demonstration units using the selected application.” That sentence creates something to compare against the request. A broad approval for an unclear future use is harder to evaluate than a defined task with a defined amount. If you cannot explain the spender’s role, put the task on hold while you verify it in official documentation.
Worked example: a finished experiment
Suppose a practice wallet originally approved a fictional token called UNIT for a fictional application called Example Swap. The allowance in the record is 900 UNIT. The planned experiment used 250 UNIT, but the reviewer has not yet checked the current onchain allowance. It would be wrong to assume that the remaining allowance must equal 650 simply by subtracting the planned use. The submitted transaction, token behavior and actual onchain state must establish the answer.
| Record field | Entry | Reason |
|---|---|---|
| Purpose | A one-time 250 UNIT experiment | Defines why access was considered. |
| Original recorded limit | 900 UNIT | A historic record, not the current allowance. |
| Application connection | Disconnected after the experiment | Closes the connection. Does not establish revocation. |
| Current allowance | Not checked | Prevents a guessed 650 UNIT value. |
| Next decision | Inspect the exact token and spender on the intended network | The review needs current evidence. |
Now suppose the correct network and spender have been independently identified, and the current allowance is displayed as 650 UNIT. The task is finished and there is no documented reason to retain that access. The reviewer can plan a revocation in the official wallet or a verified review interface, first checking the transaction details and the gas needed. The planning record should say “revocation proposed” until an executed transaction and the resulting allowance have been verified.
This distinction matters when a transaction is rejected, pending or sent on another network. A button click records intent. A transaction reference records submission. A confirmed state change is evidence of the result. Preserve those stages separately instead of overwriting the original observation with an optimistic “done.”
Use a decision rule you can explain
For this workbook, use three outcomes: keep with a recorded purpose, review before further use, or remove if the verified permission is no longer needed. These are organizational labels, not safety ratings. Keeping a permission means the user has a reason to retain it, not that the contract has been audited. Removing an allowance reduces that specific authority. It does not undo a previous transfer or eliminate unrelated permissions.
Ethereum.org recommends checking transaction details and limiting smart-contract spending permissions to the amount required for the intended task. Apply that principle to the request in front of you. A maximum token amount may still be unsuitable if the spender is wrong, the network is wrong or the authorization is not the type you expected. Amount is one field in the review, not a substitute for the rest.
| Observation | Workbook outcome | What resolves it |
|---|---|---|
| The purpose is finished and the exact allowance is known | Consider removal | Review and verify the resulting permission change. |
| The application is unfamiliar or the spender cannot be matched | Pause further use | Confirm the source and decode the request. |
| The permission supports a current, clearly defined task | Retain with a review date | Record the purpose, limit and relevant dependencies. |
| Only the app connection was removed | Review incomplete | Check remaining token authority separately. |
A signature still deserves its own review
A request that does not look like a transfer can still grant authority. Do not decide whether a signature is acceptable solely from the presence or absence of a network fee. Read the decoded fields and the official explanation of the request. If the interface only shows an opaque payload that you cannot match to the intended action, the workbook outcome is unresolved.
The useful habit is to compare intention with authority. If the task is a login, why is an asset-spending permission appearing? If the task is a small exchange, why is the request describing a different recipient or token? These discrepancies do not require a security score to notice. They require the user to stop and establish what is being requested before proceeding.
Hardware signing changes where a key signs. It does not remove this reasoning step. Compare what can be verified on the signing device with the transaction you intend. If the device cannot display a meaningful part of a complex request, record that limitation rather than treating the physical confirmation as evidence that every field was reviewed.
Keep a useful record, without secrets
A permission register should help the next review begin from evidence. Include the review date, local wallet label, network, token, spender, observed allowance, purpose, decision and verification reference. Never include a private key, recovery phrase or password. A record of where official recovery instructions can be found belongs in an operating note. The recovery secret itself does not.
Schedule another review when the use changes: an experiment ends, a team member changes role, an application changes its flow or you move to a different network. A recurring review can help catch forgotten authority, but a calendar entry is not continuous monitoring. For a specific suspicious request, reassess before signing rather than waiting for the next routine.
Download the blank permission register (CSV)
The download contains headings only. Fill it locally, and retain unknown values explicitly. Use the transfer checklist for destination and network preparation, or read the MetaMask assessment for the product context behind these permission distinctions.
Source documents
Official documentation reviewed November 11, 2026. Product terms and availability can change.