Agent Registry → Runtime Mapping

K0nsult CNC — From catalog profiles to operational workflow roles
This document defines how K0nsult maps static registry profiles to live runtime roles. Every agent profile in the registry (2,000+) is a capability blueprint. To become operational, it must be mapped to a runtime role with specific permissions, tools, and constraints. This mapping is the bridge between catalog and execution — between what an agent can do and what it is allowed to do in a given deployment.
Section 1

The Mapping Model

Every agent profile in the registry (2,000+) is a capability blueprint. To become operational, it must be mapped to a runtime role with specific permissions, tools, and constraints. The mapping pipeline ensures that no agent enters production without explicit authorization, tool scoping, data classification, and policy binding.

Registry Profile
Role Assignment
Tool Access
Policy Binding
Deployment
Runtime Instance

Registry Profile — The stored capability definition: skills, domain expertise, authority level, sector certifications.
Role Assignment — A specific operational role within a workflow (e.g., "Compliance Checker" within an audit pipeline).
Tool Access — Explicit allow/block list of tools the agent can invoke at runtime.
Policy Binding — Data classification, escalation thresholds, and autonomy caps enforced by the governance layer.
Deployment — Environment provisioning: staging vs. production, resource allocation, monitoring hooks.
Runtime Instance — The live, active agent executing tasks within its scoped permissions.

Section 2

Runtime Role Card Schema

Every agent mapped to runtime receives a Role Card — a structured record that defines its operational boundaries. The Role Card is the single source of truth for what an agent is allowed to do, access, and decide at runtime.

role_id
string · unique
Unique runtime role identifier
ROLE-TECH-K02-001
profile_source
string · registry reference
Registry profile ID this role maps from
TECH-K02
allowed_tools
array<string>
Tools the agent can invoke at runtime
[n8n_webhook, api_call, db_read]
blocked_tools
array<string>
Tools explicitly denied to this agent
[db_delete, fund_transfer, email_send]
allowed_data_class
array<string>
Data classification tiers the agent can access
[internal, public]
blocked_data_class
array<string>
Data classification tiers blocked from access
[PII, financial, regulated]
approval_rights
array<string>
Actions this agent can approve autonomously
[Level A: auto-classify, auto-route]
escalation_rights
array<string>
Conditions that trigger mandatory escalation
[confidence < 85%, amount > $1000]
sector_fit
array<string>
Target industry sectors for this role
[tech, ecommerce]
workflow_fit
array<string>
Compatible workflow types
[support_triage, content_review]
max_autonomy
enum: A | B | C | D
Highest decision level the agent can exercise
Level B
kpi_targets
object
Performance requirements measured in real-time
{accuracy: 95%, escalation_rate: <15%}
deployment_status
enum
Current operational state of this runtime role
ACTIVE / STANDBY / TESTING
Section 3

Sample Mappings — 10 Agents

The following table shows 10 representative agents mapped from their registry profiles to runtime roles. Each row demonstrates how a static profile becomes a bounded, operational instance.

Registry Profile Runtime Role Allowed Tools Data Access Max Autonomy Workflow Status
TECH-K02 Strategic Advisor api_read, report_gen internal, public Level B strategic_briefing ACTIVE
LAW-LegalRep Compliance Checker policy_lookup, doc_gen internal, regulated Level C compliance_audit ACTIVE
TECH-Grid Architecture Reviewer code_review, api_call internal Level B tech_review ACTIVE
QA-GateTurner Quality Gate test_run, report_gen internal Level A quality_check ACTIVE
COM-CommsRep Outreach Manager email_draft, crm_read internal, public Level B outreach STANDBY
AUDIT-Ledge Audit Executor db_read, report_gen, replay all (read-only) Level C full_audit ACTIVE
TECH-Titan System Architect api_call, diagram_gen internal Level B architecture TESTING
RISK-Vault Risk Assessor risk_calc, alert_gen all (read-only) Level A risk_scoring ACTIVE
MGT-Orion Program Director task_assign, status_track internal Level C program_mgmt ACTIVE
MOD-MOLT-MOD Content Moderator content_scan, flag_gen public, moderation Level B moderation ACTIVE

Click any column header to sort. Status colors: ACTIVE = live in production, STANDBY = ready but not deployed, TESTING = under validation.

Section 4

Mapping Rules Engine

The following rules are applied automatically during the registry-to-runtime mapping process. These rules enforce governance constraints before any agent becomes operational. Violations trigger a HOLD state and require human review.

Rule 1 — Process Risk HIGH Agent max_autonomy capped at Level C (advisory only, human decision required)
Rule 2 — Data class PII / financial Agent must have explicit data access grant in its Role Card. No implicit inheritance from registry profile.
Rule 3 — Sector = banking Only agents with banking_ops certification in their registry profile may be mapped to banking workflows.
Rule 4 — Workflow = customer-facing Require human_oversight flag at the output stage. No direct customer-facing output without review gate.
Rule 5 — Agent confidence < threshold Auto-escalate to next authority level. Threshold defaults to 85% but is configurable per workflow.
Rule 6 — First deployment New agents start in TESTING status for minimum 72 hours before promotion to ACTIVE.
Rule 7 — Violation count > 0 Agent must pass re-certification before runtime re-activation. Autonomy level downgraded by one tier during probation.
Section 5

Registry vs Runtime Comparison

The registry and runtime layers serve fundamentally different purposes. Understanding this distinction is critical for correct agent governance.

Aspect Registry (Catalog) Runtime (Operational)
Purpose Define capabilities Execute tasks
Count 2,000+ profiles 10–50 active per deployment
Permissions Described (declarative) Enforced (runtime policy engine)
Data Access Listed as capabilities Controlled by policy engine
KPI Defined as targets Measured in real-time
Status Model Active / Inactive ACTIVE / STANDBY / TESTING / SUSPENDED
Mutability Updated by registry admin Modified by runtime policy & governance events
Lifecycle Persistent catalog entry Ephemeral: created, suspended, terminated per deployment

Registry (Catalog)

Nature Static blueprint — describes what an agent can do
Ownership Maintained by CNC registry administrators
Update Frequency On profile change, certification, or version bump
Access Control Read: all authenticated users. Write: registry admins only
Audit Trail Version history per profile

Runtime (Operational)

Nature Live instance — defines what an agent is allowed to do now
Ownership Governed by deployment policy engine + human oversight
Update Frequency Real-time: on policy change, incident, or governance event
Access Control Scoped per Role Card — tool-level, data-level, decision-level
Audit Trail Full decision log with timestamps, escalations, and overrides
Section 6

API for Runtime Mapping

Three endpoints manage the runtime mapping lifecycle: assign a registry profile to a runtime role, query active runtime agents, and suspend a runtime agent (per-agent kill switch).

POST /api/agents/map-to-runtime
Assigns a registry profile to a runtime role. Creates a mapping record with tool permissions, data access scope, and autonomy level. Logs the mapping as a decision event for audit.
profile_id
string (required) — Registry profile ID (e.g., "TECH-K02")
role_id
string (required) — Runtime role identifier (e.g., "ROLE-TECH-K02-001")
allowed_tools
array<string> — Tools the agent can use at runtime
blocked_tools
array<string> — Tools explicitly blocked
allowed_data
array<string> — Data classification tiers accessible
max_autonomy
string — Maximum autonomy level (A, B, C, D)
workflow_fit
array<string> — Compatible workflow types
{
  "ok": true,
  "profile_id": "TECH-K02",
  "role_id": "ROLE-TECH-K02-001",
  "status": "MAPPED",
  "timestamp": "2026-03-23T14:30:00.000Z"
}
GET /api/agents/runtime-status
Returns all active runtime agents, ordered by confidence score (descending). Maximum 50 results. Used by dashboards and monitoring systems to display current operational state.
{
  "ok": true,
  "agents": [
    { "agent_id": "TECH-K02", "status": "ACTIVE", "nick": "K02", "confidence_avg": 0.94 },
    { "agent_id": "AUDIT-Ledge", "status": "ACTIVE", "nick": "Ledge", "confidence_avg": 0.91 }
  ],
  "count": 2
}
POST /api/agents/suspend-runtime
Suspends a runtime agent immediately. This is the per-agent kill switch. The agent's status is set to SUSPENDED and a decision record is logged with the reason and operator identity. Requires re-activation through the mapping endpoint to restore.
agent_id
string (required) — The agent to suspend
reason
string — Justification for suspension
suspended_by
string — Operator or system that triggered the suspension
{
  "ok": true,
  "agent_id": "TECH-K02",
  "status": "SUSPENDED",
  "timestamp": "2026-03-23T15:00:00.000Z"
}