Agent Operator
For builders using Claude Code, Codex, Cursor, Windsurf, Hermes, Devin, or Copilot to ship real projects.
- Control context and scope.
- Demand concrete artifacts.
- Use evals before trusting output.
FDE Academy is the training layer for the open-source Skill.
Field thesis: most AI projects fail because the operator lets the agent code before the problem is scoped, measured, evaluated, and owned. FDE Academy teaches the operating discipline that prevents that failure.
You will produceDomain dossier, scoping report, prototype spec, eval plan, production handoff, productization memo.
You will practiceDomain research, 6-Q interviews, ROI sizing, architecture trade-offs, eval design, runbooks.
You will avoidPrompt theater, vague requirements, eval-less demos, no-owner handoffs, SaaS-first confusion.
Choose your track
One Academy, three operator paths.
The same method serves different people. The Academy should help each user see exactly why they belong here.
FDE Consultants Protocoles
For consultants and operators who need a repeatable delivery method for client-facing AI work.
- Run discovery without vague questions.
- Translate business pain into build scope.
- Handoff systems teams can maintain.
Skill Maintainer
For contributors who want to improve the open-source Skill, templates, scripts, examples, and future MCP layer.
- Add examples and tests.
- Improve rubrics and templates.
- Productize repeated field patterns.
Curriculum
Eight modules with real deliverables.
This is not “watch videos and feel inspired.” You pass by producing artifacts that survive the FDE rubric.
00
Orientation
Install the Skill and set operator rules
Understand what the Skill is, what it is not, and how to keep agents in co-founder mode instead of autocomplete mode.
Lesson: Skill anatomyLesson: anti-patternsLab: first operator request
Pass artifact: local install plus one rejected vague prompt rewritten into an FDE request.
01
Fundamentals
Learn the 4-stage FDE loop
Scoping, prototyping, production, and feedback become the backbone of every agent-assisted engagement.
Lesson: FDE vs consultantLesson: field ownershipLab: classify engagement type
Pass artifact: one-page field operating memo with the expected outcome, owner, timeline, and artifact chain.
02
Domain research
Build the dossier before asking questions
Research market, pains, regulations, tech stack, benchmarks, recent news, and talent before discovery.
Lesson: 7 research streamsLesson: source hierarchyLab: domain dossier
Pass artifact: domain dossier with 5 insights that change the questions you ask.
03
6-Q decomposition
Turn vague AI demand into a measurable project
Force the process, output, data, error cost, current system, and success metric before allowing architecture.
Lesson: stratigraphic questionsLesson: estimating missing numbersLab: 6-Q interview
Pass artifact: completed 6-Q sheet with quantified answers and named assumptions.
04
Scoping studio
Write the report that decides GO or NO-GO
Combine stakeholder map, pain matrix, ROI, risks, architecture sketch, and 90-day plan into one decision artifact.
Lesson: stakeholder RACILesson: ROI sizingLab: scoping report
Pass artifact: 5-page scoping report using the repo template and conservative ROI threshold.
05
Prototype and scientific search
Make architecture compete before code hardens
Generate candidate architectures, validate them against held-out constraints, prune weak paths, and preserve rejected hypotheses as reusable lessons.
Lesson: stack trade-offsLesson: held-out gateLab: prototype spec
Pass artifact: prototype spec with architecture diagram, failure modes, held-out validation results, data flow, and testable hypothesis.
06
Evals
Define pass/fail before trusting the demo
Design golden cases, adversarial cases, regression checks, human review rules, metrics, and drift alerts.
Lesson: local evals vs benchmarksLesson: LLM-as-judgeLab: eval suite
Pass artifact: eval framework with target metrics, test categories, fail cases, and owner.
07
Production handoff
Move from demo to owned system
Cover deployment, rollback, observability, security, runbooks, cost projections, ownership, and incident paths.
Lesson: production checklistLesson: AI risk controlsLab: handoff doc
Pass artifact: production handoff that another team could operate tomorrow.
08
Productization
Turn field work into reusable IP
Extract repeated patterns into templates, scripts, adapters, examples, benchmarks, and eventually MCP tools.
Lesson: reusable asset ratioLesson: productization memoLab: IP extraction map
Pass artifact: productization memo ranking reusable assets by effort, ROI, and strategic fit.
12 labs that make the training real.
Lab 01Convert a vague founder prompt into an FDE request.
Lab 02Build a domain dossier for one vertical.
Lab 03Write six stratigraphic discovery questions.
Lab 04Complete a quantified 6-Q interview.
Lab 05Calculate ROI and sensitivity cases.
Lab 06Write a 5-page scoping report.
Lab 07Generate and score 4 architecture candidates.
Lab 08Draft a prototype spec with failure modes.
Lab 09Create an eval suite with happy, edge, adversarial, and regression cases.
Lab 10Write a production handoff runbook.
Lab 11Score your own work with the 6-trait rubric.
Lab 12Extract reusable IP into a productization memo.
Artifact studio
The Academy is measured by what you can ship.
Each artifact maps directly to a template or script in the Skill repository.
Domain dossier + scoping report
Industry facts, stakeholder map, 6-Q answers, pain matrix, ROI, risks, GO/NO-GO recommendation.
Repo templatesPrototype spec + eval baseline
Architecture candidates, selected stack, data flow, failure modes, eval set design, definition of done.
Eval articleProduction handoff
Deployment, rollback, observability, security, cost projections, ownership, runbook, incident path.
Handoff outputsProductization memo
Reusable IP candidates, extraction effort, product value, field insights, roadmap impact, open-source contribution path.
The 6-trait FDE rubric is the graduation bar.
Generic work fails. A passing artifact must score at least 3 on every trait, with Ownership and Decomposition at 4 or higher.
| Trait | What excellent looks like | Auto-fail signal |
|---|---|---|
| Customer Curiosity | Specific domain reality and stakeholder context. | Generic “AI can help” language. |
| Ownership | Concrete outcome, owner, date, and timeline. | “Could”, “maybe”, “explore”. |
| Decomposition | 6-Q answered with numbers. | Vague restatement of problem. |
| Empathy | Adoption, politics, maintenance, and constraints named. | Ignores the customer’s operating reality. |
| Product Sense | Shippable artifact, failure modes, and production path. | Slides, theory, or no runbook. |
| Communication | Executive summary plus technical detail. | Jargon or oversimplification. |
Capstone: ship the field package for one AI workflow.
Choose one realistic AI workflow and run the full loop. You do not need to build the entire production system to pass, but the artifact package must make implementation obvious, measurable, and ownable.
1Research the domain and write the dossier.
2Complete the 6-Q decomposition and ROI sizing.
3Write the scoping report and GO/NO-GO recommendation.
4Produce prototype spec, eval suite, and handoff plan.
5Score the package with the 6-trait rubric and revise.
Run the Academy with the Skill beside you.
Install the Skill, then use the Academy as the training path. The first prompt should not ask the agent to code. It should ask the agent to scope.
operator startlocal-first
# From the repository root mkdir -p ~/.claude/skills ln -s "$(pwd)/skill" ~/.claude/skills/fde-consultant # First Academy exercise /fde-consultant turn this vague AI idea into a 6-Q scoping interview: "We want an AI agent for customer support operations."