Skill Registry & Tool Dispatcher

Relevant source files

The following files were used as context for generating this wiki page:

• src/core/skills/skill-registry.ts
• server.ts
• src/lib/isabella-crown.ts
• src/lib/isabella-inference-engine.ts
• src/lib/creator-economy/plans.ts
• scripts/authz/isabella_scopes_catalog_v2.json
• src/data/isabellaBlueprint.ts

The Skill Registry & Tool Dispatcher manages declarative capabilities and governed execution within the Isabella Villaseñor AI ecosystem. The Skill Registry provides versioned registration, discovery, and lifecycle management for agent skills, while the Tool Dispatcher (integrated into the C.R.O.W.N. and ORION layers) executes authorized system actions within a Zero Trust sandbox. These systems decouple specific capabilities from the core orchestrator, allowing the agent to extend its functionality through registered SKILLS and TOOLS.

Sources: src/core/skills/skill-registry.ts:1-12, src/lib/isabella-crown.ts:182-195, src/data/isabellaBlueprint.ts:12-20

Skill Registry Architecture

The Skill Registry maintains a store of declarative capabilities. Each skill defines its parameters, triggers, and safety constraints. The system limits the registry to a maximum of 100 skills to maintain performance.

Data Structures

The system uses specific interfaces to define skills and their execution state.

Interface	Description	Key Fields
Skill	Defines a capability.	skillId, trigger, parameters, riskLevel, requiresConsent
SkillParameter	Defines input requirements.	name, type, required, default
SkillExecution	Tracks a specific run.	executionId, tenantId, status, input, result

Sources: src/core/skills/skill-registry.ts:14-55

Core Registry Functions

The registry provides methods to manage the skill lifecycle:

• registerSkill: Adds a new skill or updates an existing one by name. It enforces the MAX_SKILLS limit by deleting the oldest entry when capacity is exceeded.
• executeSkill: Initiates a skill run, generating a unique executionId and marking the status as running.
• completeSkillExecution: Updates an execution record with results or error messages and sets the status to completed or failed.
• enableSkill / disableSkill: Toggles the status of a skill between active and disabled.

Sources: src/core/skills/skill-registry.ts:64-159

Tool Dispatcher & Execution Sandbox

The Tool Dispatcher, primarily managed by the ORION module, handles the execution of registered tools. Execution follows a strict "Policy Gate" model where every action requires authorization from the ARGUS sentinel.

Execution Pipeline

The execution of a tool follows the canonical cognitive cycle:

The diagram shows the flow from perception to tool execution, highlighting the mandatory authorization gate.
Sources: src/lib/isabella-crown.ts:200-207, src/data/isabellaBlueprint.ts:12-20

Registered Tools Catalog

Tools are categorized by risk and module ownership. Key tools include territorial queries, voice synthesis, and cryptographic ledger commits.

Tool Key	Category	Owner	Risk Level
rdm_territory_query	Territory	ORION	Low
isabella_synthesize_voice	Synthesis	ISA	Low
argus_security_audit	Security	ARGUS	Low
sovereign_ledger_commit	Governance	BookPI	Medium
ethical_firewall	Kórima Spec	ARGUS	Critical

Sources: src/lib/isabella-crown.ts:212-230

API Endpoints

The system exposes the registry and dispatcher via a set of REST endpoints in the CROWN Gateway.

Tool Execution Endpoint

POST /api/v1/isabella/tools/execute
Executes a tool in the governed sandbox.

• Permissions: Requires tools:execute scope.
• Input: toolName (string), arguments (object).
• Behavior: Generates a traceId, audits the request, executes via executeTool, and audits the completion.

Sources: server.ts:577-610, scripts/authz/isabella_scopes_catalog_v2.json:104-111

Skill Management Endpoints

• GET /api/v1/core/skills: Lists registered skills, optionally filtered by category.
• POST /api/v1/core/skills: Registers a new skill definition.
• POST /api/v1/core/skills/:skillId/enable: Activates a specific skill.

Sources: server.ts:899-913

Governance & Security

Security for skills and tools is governed by a Zero Trust architecture.

1. Scope Validation: API keys and sessions must carry specific scopes (e.g., tools:execute, skills:manage) defined in the scopes catalog.
2. Risk Tiers: Actions are classified from Low to Critical. Critical actions, such as tools:admin, require hardware-backed step-up authentication.
3. Audit Trail: Every execution is recorded in the isabella_audit_logs table with a SHA-256 trace and provenance metadata.
4. Economic Gating: In the Creator Economy module, skill execution is metered. Users must have sufficient credits and an appropriate plan (Free, Premium, Pro, Business) to run specific skills.

Sources: scripts/authz/isabella_scopes_catalog_v2.json:104-120, src/lib/creator-economy/plans.ts:54-95, src/data/isabellaBlueprint.ts:21-27

Summary

The Skill Registry and Tool Dispatcher provide a modular framework for AI capabilities. By enforcing a mandatory Policy Gate and providing a versioned registry, Isabella ensures that all automated actions remain traceable, governed, and territorially aligned. The system balances extensibility through the Skill Registry with strict containment via the Tool Execution Sandbox.