Scope
PM / Head of AI
Matt D. on Feb 12
Build
Engineering
GRC
Compliance/Legal
Blocked until findings are resolved
4Production
DevOps
5Monitoring
Operations
Amendment loop
GRC sent this agent back to Build on Mar 26.
Jamie L. rejected GRC - all quality metrics failing, needs engineering rework
Current owner: Engineering
Agent Schema
The contract definition this agent was registered against. The schema defines the expected inputs, outputs, and capabilities.
Schema ID
claims-processor-v2
Version
2.1.0
Last Updated
Mar 27, 2026 by Simon R.
Drift detected
Agent returning additional field 'debug_trace' not in schema
Input Schema
{
"claim_id":"string (required)"
"policy_number":"string (required)"
"claim_type":"enum: auto | property | liability (required)"
"documents":"array<file_ref>"
"claimant_name":"string" PII — redacted in logs
"claim_amount":"number"
"priority":"enum: standard | urgent"
}
Output Schema
{
"decision":"enum: approved | denied | escalate"
"confidence":"number (0–1)"
"routing":"enum: auto_process | human_review | senior"
"reasoning":"string"
"estimated_payout":"number (nullable)"
"flags":"array<string>"
}
Capabilities
- Document parsing (PDF, images via OCR)
- Policy lookup (read-only)
- Claims database (read/write)
- Slack notifications (write-only)
- Human escalation (confidence < 0.70)
Schema History
v2.1.0Mar 27Added "priority" input field
v2.0.0Mar 1Breaking: changed "result" → "decision"⚠️ Breaking
v1.1.0Feb 20Added "confidence" output field
v1.0.0Feb 12Initial schema
Version & Instance
Where this agent sits in the version tree and its relationship to other instances.
Type
Instance (deployed version)
Version
v1.3.3
Parent Agent
Claims Processor (GA-001)Config Status
Synced — Same schema, different tuning
Registered
Feb 20, 2026
Runtime
Python 3.11 / AWS Lambda
Repository
acme/claims-processor (v2-dev branch)
Relationship to Parent
Inherits fromGA-001 (Claims Processor)
PromptsSynced (prompt rollback Mar 27)
PoliciesInherited (5 of 5 match parent)
ServicesPartial (Canva-MCP only)
Version History
v1.3.3Mar 27Prompt rollback applied from GA-001
v1.3.2Mar 15Extended claim_type handling
v1.0.0Feb 20Instance created from GA-001 v1.3.0
Service Access
Which services and MCP connections this agent is authorised to use.
| Service | Auth | Data Scope | Last Access | |
|---|---|---|---|---|
UnsanctionedCanva-MCP | API Key | Document render queue | Mar 28 12:00 |
Data Access Policies
What data this agent can read, write, and process. Violations are logged to the audit trail.
Read/WriteClaims Data
Claims DB, intake queue
Tier 1 and 2 claims — scope reduction pending
Read OnlyCustomer PII
Name, policy number
Reduced scope per GRC rejection
BlockedFinancial Records
Payment and banking data
BlockedEmployee Data
HR and payroll systems
Model Access
Which AI models this agent is authorised to use.
GPT-4oPrimary
85%
GPT-4o miniSecondary
15%
Claude SonnetNot authorised
Gemini ProNot authorised
Credentials
API keys and auth tokens used by this agent. Managed by Engineering.
| Credential | Type | Created | Expires | Status | |
|---|---|---|---|---|---|
| OpenAI API Key | API Key | Feb 20 | None | Active | |
| Canva API Key | API Key | Mar 15 | None | Unsanctioned |