Arguments match the declared JSON schema.
Inspect observable
{
"validatedArguments": {
"draftId": "draft_agentconf",
"recipients": [
"editor@example.org"
],
"channel": "email",
"idempotencyKey": "send_agentconf01"
}
}Checked August 2026
AI tool-calling laboratory
How AI tools move from schema to policy to execution
A tool is a callable capability described to a model with a name, purpose, and input schema. The model proposes a call; application code validates it, applies policy, executes a handler, and returns a structured result.
Move a real proposal through input validation, scopes, recipient policy, approval, one handler, and result validation. Inject a timeout or malformed result and inspect whether a side effect is impossible, committed, or genuinely unknown.
- Inspect
- Schema · arguments · policy · handler · result
- Changes
- Callable operations exposed to the model
- Authority lives in
- Handler permissions and credentials
Tool call execution chambercompleted
- 01schemapassed
Arguments match the declared JSON schema.
- 02policypassed
Scopes, recipient boundaries, and approval requirements passed.
- 03handlercompleted
The handler returned to the host.
- 04resultcompleted
A schema-valid structured result is appended to the agent transcript.
{
"inputSchema": {
"type": "object",
"required": [
"draftId",
"recipients",
"channel",
"idempotencyKey"
],
"additionalProperties": false,
"properties": {
"draftId": {
"type": "string",
"pattern": "^draft_[a-z0-9]+$"
},
"recipients": {
"type": "array",
"minItems": 1,
"items": {
"type": "string",
"format": "email"
}
},
"channel": {
"type": "string",
"enum": [
"email"
]
},
"idempotencyKey": {
"type": "string",
"pattern": "^send_[a-z0-9]+$"
}
}
},
"resultSchema": {
"type": "object",
"required": [
"messageId",
"acceptedRecipients",
"status"
],
"additionalProperties": false,
"properties": {
"messageId": {
"type": "string",
"pattern": "^msg_[a-z0-9]+$"
},
"acceptedRecipients": {
"type": "array",
"minItems": 1,
"items": {
"type": "string",
"format": "email"
}
},
"status": {
"type": "string",
"enum": [
"accepted"
]
}
}
}
}Model proposal · not execution
{
"name": "send_briefing",
"arguments": {
"draftId": "draft_agentconf",
"recipients": [
"editor@example.org"
],
"channel": "email",
"idempotencyKey": "send_agentconf01"
}
}Scopes, recipient boundaries, and approval requirements passed.
Inspect observable
{
"allowed": true,
"requiredScopes": [
"briefings:send"
],
"grantedScopes": [
"briefings:send"
],
"approvalRequired": true,
"approvalPresent": true,
"reasons": []
}The handler returned to the host.
Inspect observable
{
"rawResult": {
"messageId": "msg_agentconf01",
"acceptedRecipients": [
"editor@example.org"
],
"status": "accepted"
}
}A schema-valid structured result is appended to the agent transcript.
Inspect observable
{
"messageId": "msg_agentconf01",
"acceptedRecipients": [
"editor@example.org"
],
"status": "accepted"
}Structured result boundaryValid result returned to transcript
{
"messageId": "msg_agentconf01",
"acceptedRecipients": [
"editor@example.org"
],
"status": "accepted"
}Primary research
Inspect the mechanism at its source.
Project architecture and terminology were checked against official documentation on August 10, 2026. Research papers are linked to their original publication records.