JSON Schemas

CRP defines 10 JSON Schema files for all protocol data structures. These schemas are used for validation, documentation, and OpenAPI specification.

Schema Inventory

Schema	File	Purpose
Task Intent	task-intent.json	Input to dispatch()
Quality Report	quality-report.json	Return from dispatch()
Session Handle	session-handle.json	Return from init()
Session Status	session-status.json	Return from session_status()
Envelope Preview	envelope-preview.json	Return from preview_envelope()
Stream Event	stream-event.json	Events from dispatch_stream()
Cost Estimate	cost-estimate.json	Return from estimate_session()
CRP Error	crp-error.json	Error responses
Persisted State Header	persisted-state-header.json	Cold storage metadata
OpenAPI	openapi.json	Full sidecar API spec

All schemas are in the schemas/ directory at the repository root.

---

Task Intent

Input to dispatch(). All fields are optional:

{
  "description": "Compare React and Vue",
  "system_prompt": "You are a senior frontend architect.",
  "task_input": "Focus on performance, DX, and ecosystem.",
  "expected_output_type": "markdown",
  "max_windows": 5,
  "max_output_tokens": 4096,
  "metadata": {
    "project": "framework-eval",
    "priority": "high"
  }
}

Field	Type	Description
description	string	Human-readable task description
system_prompt	string	System prompt for the LLM
task_input	string	The actual task/question
expected_output_type	enum	text, json, markdown, code
max_windows	integer	Maximum continuation windows
max_output_tokens	integer	Max tokens per window
metadata	object	Arbitrary key-value pairs (max 50 keys)

---

Quality Report

Return from dispatch():

{
  "session_id": "abc-123",
  "window_id": "win-1",
  "output": "The complete generated text...",
  "quality_tier": "A",
  "facts_extracted": 42,
  "security_flags": {
    "injection_markers": [],
    "unicode_normalization": true,
    "control_chars_stripped": false,
    "truncation_detected": false,
    "integrity_violations": []
  }
}

Field	Type	Description
session_id	UUID	Session identifier
window_id	string	Window identifier
output	string	Unmodified LLM output (Axiom 9)
quality_tier	enum	S, A, B, C, D
facts_extracted	integer	Number of facts extracted
security_flags	object	Security check results

Quality Tiers

Tier	Meaning
S	Exceptional — fully grounded, no hallucinations
A	Strong — minor gaps
B	Acceptable — some parametric claims
C	Marginal — significant gaps
D	Poor — mostly ungrounded

---

Session Handle

Return from init():

{
  "session_id": "550e8400-e29b-41d4-a716-446655440000",
  "protocol_version": "2.0.0",
  "capabilities": ["dispatch", "ingest", "stream", "tools"],
  "created_at": "2026-04-10T12:00:00Z",
  "expires_at": "2026-04-10T14:00:00Z"
}

Field	Type	Description
session_id	UUID	Unique session identifier
protocol_version	SemVer	CRP version
capabilities	array	RBAC-granted capabilities
created_at	ISO 8601	Creation timestamp
expires_at	ISO 8601	Expiration timestamp

---

Session Status

Return from session_status():

{
  "session_id": "abc-123",
  "windows_completed": 9,
  "total_input_tokens": 34521,
  "total_output_tokens": 9836,
  "facts_in_warm_state": 372,
  "overhead_ratio": 0.061,
  "remaining_budget": null
}

---

Envelope Preview

Return from preview_envelope():

{
  "total_tokens": 3842,
  "envelope_tokens": 2100,
  "facts_included": 15,
  "saturation": 0.994
}

---

Stream Event

Events from dispatch_stream():

{
  "event_type": "token",
  "data": "The "
}

Event Type	Data	When
token	String	Each token arrives
extraction	Progress object	Background extraction
continuation	Window metadata	New window starts
window_complete	Window stats	Window finishes
done	QualityReport	All generation complete
error	Error details	Something went wrong

---

Cost Estimate

Return from estimate_session():

{
  "estimated_windows": 5,
  "estimated_input_tokens": 15000,
  "estimated_output_tokens": 10000,
  "estimated_cost_usd": 0.025,
  "confidence": "medium"
}

---

CRP Error

Error response format:

{
  "code": 1001,
  "message": "Token budget exhausted",
  "details": "Session abc-123 exceeded maximum token budget"
}

Error Codes

Code	Name	Description
1001	BudgetExhausted	Token budget exceeded
1002	RateLimit	Rate limit hit
1003	SessionExpired	Session timed out
1004	SessionLimit	Max sessions reached
1005	SessionClosed	Session already closed
1010	ValidationError	Invalid input
1011	SecurityInvariant	Security check failed
1012	SignatureInvalid	HMAC signature mismatch
1013	RBACDenied	Permission denied
1020	ProviderError	LLM provider error
1021	ProviderTimeout	LLM provider timeout
1030	StateCorrupted	Internal state corruption
1031	ChainVerificationFailed	Audit chain integrity failure

---

Persisted State Header

Cold storage / export metadata:

{
  "schema_version": "1.0",
  "protocol_version": "2.0.0",
  "created_at": "2026-04-10T12:00:00Z",
  "checksum": "blake3:abc123..."
}

Uses BLAKE3 for integrity verification.

---

OpenAPI

The full OpenAPI 3.1.0 specification for the HTTP Sidecar is available at schemas/openapi.json. Import it into:

• Postman — Import → File → openapi.json
• Swagger UI — Point to the JSON file
• Any OpenAPI client generator — Generate clients for any language