D-06 Skill Loader
Brick D: Tools, skills, plugins, MCP, connectors. French version: D-06-skill-loader
1. Role and responsibilities
Loads a skill into an agent context on demand, refusing any unsigned or blocked skill.
- Check signature, status, runtime compatibility
- Return instructions and resources
- Check declared permissions
2. Position in the architecture and interactions
- Callers: Loop Controller, Context Manager (E-09).
- Execution path: Synchronous.
- Every interaction goes through the Control Plane API v1 (
/v1) behind the API Gateway (D-14).
3. Open source component and developed part
- Open source: none (to develop)
- Developed: Fully developed (progressive loading).
- External integration through an adapter
eai/integrations/<system>.pywith an in-process fallback when the URL is not configured.
4. Data model
Tables (schema eai, tenant_id RLS):
eai.skills
Common columns: id uuid, tenant_id uuid NOT NULL, created_at, updated_at; RLS policy current_setting('app.tenant_id', true)::uuid.
5. Contract
- Applied contract: skill-manifest
- JSON Schema:
contracts/v1/skill-manifest.schema.json - Key fields:
spec.permissions, spec.signature, spec.compatible_runtimes - Validation:
POST /v1/contracts/{name}/validate; manifest apply:POST /v1/manifests/apply.
6. API endpoints
GET /v1/skills/{name}/load
Request GET /v1/skills/{name}/load:
{}
Response:
{
"name": "invoice-reader",
"version": "1.0.3",
"instructions": "# Invoice reader...",
"resources": [
"templates/invoice.json"
],
"digest": "sha256:ab12..."
}
7. SDK examples
Python:
from enterprise_ai import EnterpriseAI
client = EnterpriseAI(base_url="<control-plane-url>", api_key="eai_xxx_secret")
skill = client.skills.load("invoice-reader")
TypeScript:
import { EnterpriseAI } from "@enterprise-ai/sdk";
const client = new EnterpriseAI({ baseUrl: "<control-plane-url>", apiKey: process.env.EAI_API_KEY });
const skill = await client.skills.load("invoice-reader");
Java:
EnterpriseAIClient client = EnterpriseAIClient.builder()
.baseUrl("<control-plane-url>")
.apiKey(System.getenv("EAI_API_KEY"))
.build();
Skill skill = client.skills().load("invoice-reader");
Go:
c, _ := eai.NewClient(eai.WithBaseURL("<control-plane-url>"), eai.WithAPIKey(os.Getenv("EAI_API_KEY")))
skill, _ := c.Skills.Load(ctx, "invoice-reader")
8. Configuration
- Environment variables:
EAI_DATABASE_URL - Helm values keys:
sandbox.enabled,controlPlane.config
9. Security controls
- Refusal 403 skill_unsigned or skill_blocked
- Every request carries the tenant resolved from the credential; PostgreSQL RLS
- Every mutation and policy decision writes a hash-chained audit event
10. Observability
- Prometheus metrics:
eai_skill_loads_total{result} - Traces: OpenTelemetry spans with GenAI semantic conventions (
gen_ai.operation.name,gen_ai.agent.id,gen_ai.tool.name), attributeseai.tenant_id,eai.run_id,traceparentpropagation. - Audit actions:
skill.load
11. FinOps units
- Instruction tokens counted in the context.
- Emitted as Usage Event (
POST /v1/finops/usage) with tenant, project, agent and run.
12. Failure modes and resilience
- Cache by digest, invalidated on block
- Normalized errors (error envelope, message_i18n en/fr)
13. Standalone adoption (client of a client)
- enterprise_ai.skills.
- SDK:
enterprise_ai.tools / mcp / skills / plugins / connectors / channels - Minimal deployment: Helm chart with
controlPlane.enabled=trueand PostgreSQL; disabled external components use their in-process fallback.
14. Acceptance tests
- Unsigned skill -> 403
- Blocked skill: refused on next load
- Cross-tenant access refused
- Error returned in fr with Accept-Language: fr