Start here. Almost every failure is one of eight things.
SymptomWhat it meansWhat to do
401 on every callThe key is missing, revoked, or not a skills keyCreate a fresh key and paste it in again
Blocked before anything ranA spend cap refused the callRaise the cap or ask for less
”Cannot be priced”The tool bills by time or tokens, so the cost is unknowable up frontPick a tool with a fixed or per-result price
The run stops at 120 secondsThe request hit its ceiling, not the runRe-request the run by id, or use wait: false
A credential is named and missingThe tool needs a key nobody holds for youUse another tool for the same job, or tell us
Refused as against a vendor’s rulesThe request would breach an acceptable-use policyRephrase the task, or use a different source
Empty results, no errorA valid call with too narrow a queryWiden the query before raising the limit
The agent ignores the skillIt never fetched the filePaste the setup line again in a fresh session

​
A 401

Three causes, in order of likelihood: the key was pasted with a character missing, the key was deleted in the app, or the request is going somewhere a skills key does not reach. A key covers the tool catalog and your own runs. It cannot sign in, read your chats or files, change your plan or lift your caps — so a 401 on anything else is the scope working. Create and revoke keys at agentmore.app/app/api-keys, and see API keys. Rotating: paste the new key in first, then delete the old one.

​
A call refused by a cap

Nothing was sent to a vendor and nothing was charged. A refusal is terminal — the call does not retry itself into your cap, it stops and tells you which limit it hit. Your two caps are per call and per day, at agentmore.app/app/budget. They sit under our own ceilings, so you can tighten but never raise past ours. The cheapest fix is usually the query, not the cap: ask for 20 results instead of 200. Detail: Spend caps.

​
A call that cannot be priced

Every call is priced from the tool’s real price shape before it runs. Two shapes cannot be priced from the input at all — anything billed by duration or by tokens consumed. Those calls are refused rather than run at an unknown cost. This is not a bug to work around. Pick a tool priced per call or per result; discovery shows the shape next to the price.

​
A run that times out

A request waits up to 120 seconds. That is the ceiling on the request, not on the work — the run carries on afterwards. If you are calling over HTTP, send wait: false, keep the runId, and poll GET /api/supertool/runs/:id. If you are in chat or in your own agent, ask again for the same result rather than starting the job over; a second start is a second charge.

​
A tool that needs a credential we do not hold

Some catalogued tools sit behind a vendor account that is not part of the managed set. The failure names the exact credential it wanted and stops — it does not silently try something else. Look for another tool covering the same platform; the catalog usually has several. If nothing does, mail rick@agentmore.app with the tool id.

​
A request refused on acceptable use

Some requests are refused because running them would breach a vendor’s acceptable-use policy. The vendor accounts are shared, so one breach affects every account using them. Nothing is charged. Rephrase the task in terms of what you actually need — public company information, published posts, your own accounts — and it usually runs.

​
Still stuck

Check the call log on your dashboard. Every call is there with its status, its cost and the reason it stopped, which is faster than reproducing the failure. Then mail rick@agentmore.app with the tool id and roughly when it ran.