Skip to main content
Hyde Workshop Human-governed AI

A.M.A.N.D.A. // Supervised decision runtime

Six governed stages. One accountable decision.

A.M.A.N.D.A. analysis is Hyde Workshop's supervised decision method. It defines the mission, verifies evidence, audits the implementation, routes the work, tests defenses, and prepares a recommendation that stops at a human approval gate.

Explore the six stages Build a planning brief
AI Astra, the accountability agent of the A.M.A.N.D.A. supervised decision method

Each stage owns one responsibility. No stage owns the final decision.

AI Amy, the Adaptive Mission Agent

Adaptive Mission Agent

AI Amy

Turns an unclear request into a defined mission with audience, objective, scope, constraints, acceptance criteria, and a named human owner.

Primary output: mission brief and acceptance criteria.

AI Maya, the Multi-Source Mapping Agent

Multi-Source Mapping Agent

AI Maya

Maps grounded evidence, official documentation, dates, versions, claims, missing information, and source quality before any conclusion is formed.

Primary output: evidence map and source-grounding record.

AI Aria, the Analytical Audit Agent

Analytical Audit Agent

AI Aria

Audits technical quality, accessibility, SEO, performance, compatibility, maintainability, and whether a claimed improvement can actually be demonstrated.

Primary output: audit findings and remediation requirements.

AI Nancy, the Neural Navigation Agent

Neural Navigation Agent

AI Nancy

Maps the implementation route across page code, Elementor, WordPress, plugins, hosting, caching, analytics, consent, dependencies, handoffs, and rollback.

Primary output: controlled implementation route and sequencing.

AI Dawn, the Defense Diagnostics Agent

Defense Diagnostics Agent

AI Dawn

Examines security, privacy, permissions, consent, unsafe actions, sensitive data, backups, recovery, incident response, and rollback requirements.

Primary output: risk register, safeguards, and recovery requirements.

AI Astra, the Accountability Agent

Accountability Agent

AI Astra

Combines the earlier stages into a recommendation for human review. Astra can prepare a decision, but cannot grant human approval or authorize autonomous deployment.

Primary output: READY FOR HUMAN REVIEW, or a governed stop or revision state.

How an A.M.A.N.D.A. analysis runs

An A.M.A.N.D.A. analysis is a six-step review that ends with a decision made by a named person. One role handles each step, and each role leaves a written artifact behind, so the reasoning can be checked, challenged, or reversed.

When to use it

  • Adopting a new AI tool, or connecting an agent to files, accounts, or APIs.
  • Changing a production WordPress page, template, or plugin.
  • Publishing AI-assisted content that makes factual claims.
  • Automating a task that spends money, sends messages, or deletes data.

What you end up with

A decision record: the six artifacts, the recommendation (adopt, pilot, restrict, or reject), the name of the person who decided, and the date.

What it does not do

It does not deploy anything, approve its own recommendation, or guarantee an outcome. It does not replace legal, security, or compliance advice. Astra can prepare a decision, but only a person can make it.

Illustrative example: adding an AI chat widget to a WordPress site

  1. Amy: goal is to answer pre-sales questions, with no pricing promises, and the site owner decides.
  2. Maya: read the vendor's documentation, data-handling terms, and current version, and note anything unverified.
  3. Aria: measure page speed and accessibility with and without the widget on a staging copy.
  4. Nancy: plan the order of work: staging, one page, then sitewide, with cache-purge steps.
  5. Dawn: check what visitor data the widget sends, where API keys live, and how to switch it off fast.
  6. Astra: brief the owner to pilot on one page with a monitoring window, or reject if speed or privacy checks fail.

This example is illustrative, not a case study. For a public list of common risks in applications built on large language models, see the OWASP GenAI Security Project (opens in a new tab).

Browser-only brief builder.

Describe the outcome and get a structured route, not a black-box promise. It updates as you type and nothing leaves your browser.

Do not enter passwords, API keys, or customer data.

WordPress / Elementor controlled repair

Primary constraint: risk and safety. Separate page-level defects from Elementor, WordPress, plugins, hosting, caching, analytics, and consent before changing production.

  1. Amy: missionDefine the defect, affected users, desired result, acceptance criteria, exclusions, and human owner.
  2. Maya: evidenceCollect complete code, screenshots, template context, plugins, versions, cache layers, and relevant documentation.
  3. Aria: auditAudit HTML, CSS, JavaScript, headings, IDs, ARIA, images, SEO, accessibility, performance, and compatibility.
  4. Nancy: routingMap the implementation across page code, Elementor, WordPress, plugins, hosting, cache and CDN, and analytics and consent.
  5. Dawn: defenseCheck credentials, scripts, forms, external dependencies, backups, recovery, safeguards, and rollback.
  6. Astra: decisionPrepare the recommendation and stop at HUMAN_REVIEW before publishing or deployment.

Next: turn a workflow into controls in AI Forge, watch the sequence in the Workshop, or send a project brief.