This documents the public skill/ product: installation, runtimes, usage, output contracts, scripts, and contribution. The hosted MCP layer stays marked Bêta.
Field thesis: install once as a Claude Code plugin, or symlink skill/ manually. No hosted account is required for the open-source Skill.
# Claude Code — one command (skill + Modex MCP) /plugin marketplace add selectess/fde-consultants-protocoles /plugin install fde-consultant@fde-consultant # then ask your agent /fde-consultant scope this new AI project idea
# Or symlink manually from the repository root
mkdir -p ~/.claude/skills
ln -s "$(pwd)/skill" ~/.claude/skills/fde-consultant
Every FDE deliverable ends with an FDE Assurance Score (0–100, DeepSCR-verified). Browse the public, hash-chained Assurance Registry.
Local MCP server (free) — the Skill ships a 7-tool MCP server (run python3 -m mcp_server from skill/): fde_recon · fde_decompose · fde_roi · fde_scientific_search · fde_evals · fde_ontology · fde_trust_score.
| Runtime | Use today | Future path |
|---|---|---|
| Claude Code | Install skill/ into ~/.claude/skills/fde-consultant. | Native skill workflow + community examples. |
| Codex | Auto-scan ~/.agents/skills/, or load SKILL.md as project guidance. | Packaged skill distribution. |
| Cursor / Windsurf | Add skill/ as project context and invoke the role explicitly. | MCP wrapper as the hosted Bêta matures. |
| Hermes / open agents | Use the Markdown entrypoint + scripts/templates as callable local assets. | Adapters for skill loaders and local tool execution. |
Trigger the Skill when the work needs a real delivery artifact: scoping report, prototype spec, architecture, eval framework, production handoff, productization memo, or roadmap.
/fde-consultant scope a customer-support AI triage project /fde-consultant architect a production RAG system for legal docs /fde-consultant evaluate this prototype handoff /fde-consultant productize the reusable parts of this engagement
| Path | Purpose | Why it matters |
|---|---|---|
skill/SKILL.md | Activation rules, loop, anti-patterns, scoring. | The agent entrypoint. |
skill/references/ | FDE, AI agents, SaaS, business AI, evals, benchmarks. | Deep context without bloating every prompt. |
skill/prompts/ | Domain research, discovery interview, strategic questions. | Reliable scoping behavior. |
skill/scripts/ | 6-Q validator, ROI calculator, ontology extractor, eval runner, scientific search. | Executable rigor, not pure text. |
skill/templates/ | Scoping, prototype, production handoff, productization memo. | Consistent output contracts. |
skill/examples/ | Multi-industry examples. | Fast onboarding and proof. |
Executive summary, stakeholder map, 6-Q answers, pain matrix, ROI, constraints, recommendation.
Candidate architecture, selected stack, eval baseline, integration plan, failure modes.
Runbook, security, observability, deployment, cost, ownership, incident paths.
Reusable IP candidates, extraction effort, ROI, strategic fit, field insights.
scripts/decompose_problem.py — validates whether the problem is concrete enough to build.scripts/roi_calculator.py — annual impact, payback, NPV, and sensitivity.scripts/ontology_extractor.py — extracts actors, systems, processes, objects, metrics.scripts/evals_runner.py — scores deliverables against the 6-trait FDE rubric.scripts/scientific_search.py — candidate architecture refinement before promotion.python3 -m pytest skill/tests -q
Every FDE output passes six traits: customer curiosity, ownership, decomposition, empathy, product sense, communication. Auto-reject generic AI advice, slide-only deliverables, no numbers, no evals, no owner, or no production path.
The best contributions add proof: sharper examples and templates, new benchmark references, platform adapters, tests, bug fixes, and real field lessons. The method should compound with every serious use. Open a PR on GitHub →