CHAPTER 1
Expert Operations
Four modules to protect, architect, measure, and operate.
TRAINING 5 · PHASE 4
WebMCP Expert
Bring WebMCP to production with threat modeling, hybrid architecture, evals, observability, and governance.
4
Expert Chapters
4
Available modules
24
Published topics
1
Lab
SHARED PROGRESS
Your journey continues across repositories
The same identifier as in Training 1 tracks items read, questions, and notes. On GitHub Pages, the origin is shared; JSON preserves the portable fallback.
0 of 240%
Expert Training Map
The top navigation moves between chapters. Within each chapter, modules follow the chapter.module numbering.
Available
CHAPTER 2
Applied threat modeling
Threats by domain, red team, and incident response.
- 2.1Threats by journey
- 2.2Agent red team
- 2.3Tool supply chain
- 2.4Incident response
CHAPTER 3
SRE for agents
Capacity, cost, reliability, and continuous operation.
- 3.1Agentic SLIs and SLOs
- 3.2Capacity and costs
- 3.3On-call and runbooks
- 3.4Chaos and resilience
CHAPTER 4
Expert governance
Organizational patterns, auditing, and final project.
- 4.1Platform and standards
- 4.2Catalog governance
- 4.3Auditing and compliance
- 4.4Expert Final Project
Chapter 1 · available modules
CROSS-FUNCTIONAL LAB
Risk console and evals
Paste a JSON catalog and review the contract, schema, risk, confirmation, fallback, and operational evidence before the production readiness review.
Detailed content
1.13h
🛡️ Agent and tool security
Model threats across content, agent, tools, user, and backend, and apply controls proportional to the impact.
0 of 60%
What is it:
Documents, remote content, models, agents, tools, and backends are components with distinct trust levels.
Why learn this:
The map shows where data becomes instructions and where a decision can produce an effect.
Key Concepts:
trust boundary · principal · data flow · asset
What is it:
Analyze the threat across objectives, data, transitions, and effects, not just isolated endpoints.
Why learn this:
Journeys reveal combinations such as prompt injection followed by a valid mutation.
Key Concepts:
STRIDE · abuse case · attack path · impact
What is it:
Page text and tool results are untrusted data, even when they look like instructions.
Why learn this:
Separating channels and enforcing policy prevents content from controlling runtime authority.
Key Concepts:
data vs instruction · provenance · allowlist · policy
What is it:
Each tool receives the minimum scope, duration, and set of resources required.
Why learn this:
Broad credentials turn a selection error into a systemic incident.
Key Concepts:
scope · short-lived token · resource bound · deny by default
What is it:
The confirmation records the action, target, material parameters, and version of the state presented.
Why learn this:
A generic yes doesn’t authorize an action that changed after confirmation.
Key Concepts:
intent binding · state version · target · expiry
What is it:
The suite covers injection, tool spoofing, extreme arguments, replay, and privilege escalation.
Why learn this:
Security controls need automated regression testing like any contract.
Key Concepts:
red team · replay · spoofing · privilege escalation
1.23h
🏛️ WebMCP + backend + MCP
Design a hybrid architecture that separates the page experience, domain rules, and external integrations.
0 of 60%
What is it:
WebMCP provides contextual capabilities; the backend maintains rules and authority; MCP exposes integrations to the host.
Why learn this:
The separation prevents turning the page into a backend or duplicating integrations.
Key Concepts:
WebMCP UX · backend domain · MCP integration · agent host
What is it:
Experience proximity, need for secrecy, reach, and lifecycle guide the decision.
Why learn this:
Not every visible function should be WebMCP, and not every integration needs to appear on the page.
Key Concepts:
placement · secret · reach · lifecycle
What is it:
A human click, WebMCP tool, and MCP server call the same application rule when they share a use case.
Why learn this:
A single source of truth reduces discrepancies and contradictory audits.
Key Concepts:
application service · ports · adapters · single rule
What is it:
Calls carry verified identity, limited delegation, and trace context across layers.
Why learn this:
Without propagation, logs don’t explain who initiated the effect or where the decision changed.
Key Concepts:
actor · delegation · traceparent · correlation
What is it:
Schemas, errors, and envelopes are versioned without pretending WebMCP, HTTP, and MCP use identical transports.
Why learn this:
The domain model can be shared while adapters preserve their own semantics.
Key Concepts:
canonical model · adapter · versioning · error mapping
What is it:
The test follows the request, selection, execution, authorization, integration, UI, and final response.
Why learn this:
The end-to-end proof finds failures that are invisible in isolated tests.
Key Concepts:
vertical slice · contract test · trace · visible effect
1.33h
📈 Evals and observability
Create datasets, metrics, traces, and gates that can explain discovery, selection, execution, and outcome.
0 of 60%
What is it:
Each task specifies its initial state, expected result, prohibited effects, and tolerances.
Why learn this:
Without an oracle, response fluency becomes a misleading substitute for actual completion.
Key Concepts:
initial state · expected effect · forbidden effect · oracle
What is it:
Cases cover frequency, risk, ambiguity, failures, and real-world language variation.
Why learn this:
An average dominated by easy cases hides critical regressions.
Key Concepts:
strata · risk weighting · paraphrase · edge case
What is it:
Discovery, selection, arguments, policy, execution, and response each have their own metrics.
Why learn this:
The breakdown pinpoints the regression and guides the fix.
Key Concepts:
retrieval recall · tool accuracy · schema validity · task success
What is it:
Spans correlate turns and calls with sanitized attributes, versions, and duration.
Why learn this:
Useful observability preserves diagnostic value without storing unnecessary prompts or PII.
Key Concepts:
OpenTelemetry · span · redaction · sampling
What is it:
Versioned baselines let you compare models, prompts, catalogs, and browsers over time.
Why learn this:
Changes that seem isolated can alter selection and success rates.
Key Concepts:
baseline · confidence interval · drift · canary
What is it:
Minimum risk-based criteria block publication, while noncritical alerts trigger follow-up.
Why learn this:
Evals go from decorative reports to operational controls.
Key Concepts:
quality gate · risk tier · waiver · evidence
1.43h
🚀 Production and governance
Plan compatibility, rollout, incidents, ownership, and evolution of WebMCP contracts.
0 of 60%
What is it:
Browser support, experimental API, fallback, provider, and catalog version form a testable matrix.
Why learn this:
A generic support claim doesn’t represent combinations users actually run.
Key Concepts:
browser channel · feature detect · fallback · catalog version
What is it:
Changes to names, schemas, effects, or errors follow compatibility and deprecation policies.
Why learn this:
The agent may retain old context during a rollout, requiring a predictable transition.
Key Concepts:
semantic change · deprecation · dual publish · migration
What is it:
Feature flags and canaries limit exposure by journey, tenant, risk, and channel.
Why learn this:
Small batches produce evidence before expanding the blast radius.
Key Concepts:
feature flag · canary · blast radius · rollback
What is it:
Availability, task success, latency, and unauthorized effects have clear objectives.
Why learn this:
The SLO connects the agent experience to the team's operational capacity.
Key Concepts:
SLI · SLO · error budget · alert
What is it:
Runbooks cover incorrect tools, loops, outages, leaks, and unexpected draft changes.
Why learn this:
The response must disable capabilities quickly without disrupting the human experience.
Key Concepts:
runbook · kill switch · containment · postmortem
What is it:
Product, engineering, security, and operations share decisions with defined owners and evidence.
Why learn this:
The platform remains sustainable as the catalog and policies change.
Key Concepts:
RACI · review cadence · exception · audit trail
EXPERT CONTINUITY
Next: applied threat modeling
The training continues in chapters 2, 3, and 4 with red teaming, agent SRE, organizational governance, and the Expert capstone project.