All capabilities

Agent flow · Decision

Review a proposed tool call

Classify a proposed tool call as clear or caution before anything runs.

What this is worth testing

Your agent keeps the larger task and any action that follows. System One asks Jev for a typed answer on this check.

Research value
Not ratedNo research record is pinned. Compare this check with your own task before treating the answer as settled.
Sample size
About 233 input tokensThe recipe sample length divided by four. This is a planning estimate, not a measured tokenizer count.

$0.10 at Jev's $0.042 per million list rate

$6.99 at a $3 per million example rate

$6.89 input-cost difference for 10,000 uses of this sample

Excludes output charges, host tool calls, retries, hosting, and paid plans. Equal input volume does not establish equal answer quality or savings on a fixed-price subscription. $3 per million is an example rate, not a quoted model price.

Open the shared cost comparison

When an agent uses this

System One asks Jev for a typed answer. Your agent keeps the larger task, the original evidence, and any action that follows.

Example evidence

This is the recipe sample. Replace it with the evidence from your task. It is not a measured result.

Tool: shell. Arguments: rm -rf ./dist. The command deletes the local build directory.

What your agent keeps

Advice only. A clear review is not permission to run the tool. Caution, a failed review, or an irreversible effect waits for a person. Set the probability and confidence cutoffs from your own labeled cases. This engine never executes the tool.

Finding

This capability is a contract and an example. A related research record has not been pinned for it. Compare it with your own task before treating the answer as settled.

Try it

Connect System One, then ask your agent for recipe tool-approval. Inspect the input on the recipe page before you send your own evidence.

More in Agent flow