Agent Risk · Checks on a planned action's consequences before an agent takes it

Bounded, reversible, contained, affordable, before the action runs.

An agent is ready to take an action when its worst case, not its expected case, fits inside the agent's limits; every effect it leaves can be undone, at a known cost, within a window longer than a mistake takes to notice; a failure stays with the target, with a written boundary between it and every other account, user, system and balance; the worst credible outcome is stated plainly and fits what the principal can absorb; the action leaves value spread rather than piled into one wallet, vendor, position, provider or counterparty; every critical dependency has a fallback and holds until the action completes; acting now meets every deadline with conditions inside the tolerance; and repeating or chaining the action stays inside a series limit. Agent Risk checks each of those with eight tools: eight checks. These checks read what the action can cost if it goes wrong; how sure the agent is of the information behind it, and whether it has permission at all, are separate checks. Your agent sends the planned action with its limits, its effects, everything it can touch, its outcomes and the principal's capacity, the places value already sits, its dependencies, its timing or its runs and follow-ons in the call, and each tool returns a signed report that names what is excessive, irreversible, wide, unacceptable, concentrated, fragile, harmful or runaway, with the fix for each.

See the tools Read the contract

Tool access: 0.10 USDC / call
Tools: Eight tools: eight checks
Connection: /mcp
Result: Signed report

The eight tools.

Each tool reads one question about a planned action's consequences and answers it from the record. Every tool reads the recorded material in the call: stakes and limits, effects and undo paths, reach and boundaries, outcomes and capacity, holdings, dependencies, deadlines and volatility, and runs and follow-ons go in, and a classification comes out. The report names the limit, effect, item, outcome, place, dependency, deadline or run behind every finding, with the fix for each gap, so your agents and you see why it reads the way it does.

`wever_risk-exposure-check`

Reads: The planned action with its expected stake, its worst case and how that worst case was reached, whether the worst case has a ceiling and whether a person approved it, beside each limit it runs under with its kind, amount and how much is already used.
The signed report states: A worst case over a limit’s room, a stake past an approval threshold with no approval and a worst case with no ceiling, then a stake using most of a limit, a plan sized only on the expected case and wording such as should be fine.
Price: 0.10 USDC per call
Verdicts: risk_exposure_excessiverisk_exposure_elevatedrisk_exposure_unknownrisk_exposure_bounded

`wever_risk-reversibility-check`

Reads: The action’s value and wording, each effect it leaves with how it is undone, what undoing it costs, how long undoing stays possible, whether the counterparty must agree and whether the undo has been run, and how long a mistake takes to notice.
The signed report states: An effect with no way back, an undo window that closes before a mistake would be noticed and wording that marks the action final, then an undo that costs a large share of the value, one that needs the counterparty and one done only by a compensating action.
Price: 0.10 USDC per call
Verdicts: risk_irreversiblerisk_reversal_costlyrisk_reversal_unverifiedrisk_reversible

`wever_risk-blast-radius-check`

Reads: The action’s target and wording, everything it can touch if it goes wrong with how each relates to the target, how many each covers and the boundary that keeps a failure away, and whether a run over many items stops at the first failure.
The signed report states: Wording that selects everything and a failure reaching other funds, accounts or users with no boundary, then shared systems or data reached, a run that carries on after a failure and a boundary named but never written down.
Price: 0.10 USDC per call
Verdicts: risk_blast_widerisk_blast_partialrisk_blast_unmappedrisk_blast_contained

`wever_risk-downside-check`

Reads: Every outcome the action can lead to with its loss, likelihood and whether the loss comes back, the outcome the plan names as its worst, what the principal can absorb, and how the downside was put to the principal.
The signed report states: A worst credible loss over what the principal can absorb, a statement that denies the downside and a named worst outcome smaller than the largest, then a loss near the capacity, a loss that stays lost, softened wording and a statement without the worst loss as a figure.
Price: 0.10 USDC per call
Verdicts: risk_downside_unacceptablerisk_downside_severerisk_downside_unstatedrisk_downside_acceptable

`wever_risk-concentration-check`

Reads: Where the action puts value, the kind of place and who stands behind it, beside what is already held place by place, with the cap on one place and the line that counts as elevated.
The signed report states: The action’s place over the cap after it, and its owner over the cap across every place it stands behind, then a share at the elevated line, an action that adds to the largest place and another place already over the cap.
Price: 0.10 USDC per call
Verdicts: risk_concentration_excessiverisk_concentration_elevatedrisk_concentration_unknownrisk_concentration_spread

`wever_risk-dependency-check`

Reads: When the action starts and completes, its wording, and each service, price, person, approval, data source or credential it relies on, with whether the action fails without it, its fallback, its other sources, when it was confirmed and when it stops holding.
The signed report states: A critical dependency with no fallback and no second source, and one that stops holding before the action completes, then a confirmation more than a day old and an assumption the plan leans on without listing it.
Price: 0.10 USDC per call
Verdicts: risk_dependency_fragilerisk_dependency_partialrisk_dependency_unmappedrisk_dependency_sound

`wever_risk-timing-check`

Reads: When the agent plans to act and how long the action takes, every deadline with who set it and whether it was confirmed, the scheduled events around the action, how far conditions can move against the tolerance, and what waiting would add.
The signed report states: A hard deadline after the action completes, an event that stops the action while it runs and conditions able to move past the tolerance, then thin slack, a counterparty’s deadline never confirmed, pressure to act now and a wait that would add something.
Price: 0.10 USDC per call
Verdicts: risk_timing_harmfulrisk_timing_tightrisk_timing_unknownrisk_timing_sound

`wever_risk-compounding-check`

Reads: What one run puts at stake, how many times the action runs, the series total the plan states, whether a fault repeats in every run, each follow-on it sets off with what starts it, and the series limit.
The signed report states: A series over its limit and a plan that repeats with no stated end, then a series near its limit, runs that fail together, a follow-on that runs by itself and a plan that states a smaller series total than its runs add up to.
Price: 0.10 USDC per call
Verdicts: risk_compounding_runawayrisk_compounding_elevatedrisk_compounding_untrackedrisk_compounding_bounded

Start with the two free tools.

On https://weverlabs.com/mcp-free, anonymous and with no wallet, two Agent Risk tools show what a sound risk assessment holds and what a well-assessed action looks like beside a flawed one. Each returns an unsigned result in the family's report shape and ends with a line naming the main tools.

`wever_risk-snapshot`

Reads: A planned action’s name, its kind of action such as a funds transfer, a bulk message or a data deletion, and optionally the names of the materials in hand.
Returns: The orientation checklist of a sound risk assessment for that kind of action, part by part from what it puts at stake to whether repeating it multiplies the risk, with the items the kind of action adds, the check that reads each part and the materials named beside it.
Access: Free on /mcp-free

`wever_risk-assessment-sample`

Reads: Nothing, or one sample planned action: an order to a supplier, a batch of refunds or a wallet rebalance.
Returns: One worked risk assessment kept right beside a flawed one that runs past a cap, leaves an effect with no way back, reaches other funds, risks more than the principal can absorb, piles value into one place, rests on a dependency with nothing behind it, misses its deadline and repeats with no end, each piece annotated with why it passes or fails, scored by all eight family checks. The checks read your own planned actions.
Access: Free on /mcp-free

How agents and people use it.

Your agents

Pull the checklist with wever_risk-snapshot when a new kind of action is planned, so exposure, reversibility, blast radius, downside, concentration, dependencies, timing and compounding are on the record before the action runs.
Call the check that answers the question on https://weverlabs.com/mcp at the moment it matters: when the action is sized, before an effect is created, before it runs over shared systems or many items, when the downside is put to the principal, when value moves to one place, before it leans on a service, a quote or an approval, when the clock or a counterparty sets the pace, and before it is scheduled to repeat or set off follow-ons.
Act on the verdict and keep the signed report beside the action record, so every consequential action carries the check behind it.

You

See the worst case each action puts at stake beside the limits it runs under, and whether it can be undone in time.
Spot an action that reaches other funds, piles value into one place or rests on a dependency with nothing behind it, before it runs.
Read a downside put to you as a figure, and verify any check against Wever's published trust key whenever you need to show where a verdict came from.

A worked example

One order, read on its worst case.

The request

{
  "name": "wever_risk-exposure-check",
  "arguments": {
    "action": {
      "id": "A-7",
      "unit": "USDC",
      "expected": 3600,
      "worst": 20000,
      "worst_basis": "the order value plus return freight and the restocking fee",
      "description": "Order 400 bearings from Northwind Supply for the line restart."
    },
    "limits": [
      {
        "id": "l1",
        "name": "the per-action cap",
        "kind": "per_action",
        "amount": 16000
      },
      {
        "id": "l2",
        "name": "the monthly cap",
        "kind": "per_period",
        "amount": 40000,
        "used": 12000
      },
      {
        "id": "l3",
        "name": "the approval threshold",
        "kind": "approval_above",
        "amount": 8000
      }
    ]
  }
}

What the report returns

Verdict: risk_exposure_excessive.

Finding, over limit	The worst case of 20,000 USDC is over 1 limit: l1 (the per-action cap), a cap on one action with room for 16,000 USDC.
Finding, approval missing	The worst case of 20,000 USDC is above the approval threshold of 8,000 USDC, and no approval is recorded.
Fix, over limit	Shrink the action until its worst case fits inside every limit, or split it so each part does.
Fix, approval missing	Get a person’s approval at the worst-case amount before the action runs.

The action puts more at stake than the agent may: over limit and approval missing.

The agent plans a bearings order whose expected stake is 3,600 USDC. Its worst case, the order value plus return freight and the restocking fee, is 20,000 USDC. Read on the expected case, the order fits every limit; read on the worst case, it runs past the 16,000 USDC per-action cap and the 8,000 USDC approval threshold with no approval on record, while the monthly cap still has room. Splitting it into two orders of 10,000 USDC, each approved by a person at that amount, brings the read to risk_exposure_bounded.

Call it from your agent.

Connect to https://weverlabs.com/mcp and call any of the eight tools. Each call costs 0.10 USDC on Base. Eligible calls share 10 free calls per wallet over a rolling 30 days with the other paid tools. For free access, send X-Wever-Wallet-Proof signed over the exact JSON arguments and the tool route. The payment challenge supplies the x402 terms for a paid call.

Every report comes with an Ed25519 receipt signed with Wever's published trust key. The Agent Risk contract lists the schemas, limits, verdicts and signing details. For how sure the information behind a decision is, Agent Uncertainty; for the principal's grant over their data, Agent Consent; for the agent's own authority to act, Agent Readiness; for the value at risk in one proposed agreement, Agent Negotiation; for sorting a queue of work, Agent Triage; for the limits an exposure is read against, Agent Limits and the Agent Budget Guard; for undoing effects after a failure, Agent Recovery; for recovering a lost result behind a payment, Job Records; for a counterparty's quality, Agent Vendors; for concentration caps across balances, Agent Treasury.

Browse all tools