Catalog
A reusable library of AI expertise for business processes.
The catalog is the expertise layer of the system. It turns business methods and technical knowledge into structured, versioned, reusable capabilities — so expertise can be assembled into controlled workflows instead of being rebuilt from scratch for every project.
It currently packages 38 specialized agents across 37 skill folders — 425 individual skills in all — covering software engineering, AI, product management, Agile, quality, consulting, project management and HR. Each agent has a defined role, a focused area of expertise and reusable skills; agents combine inside structured workflows and run through the Agentic Runtime. The catalog is versioned and machine-readable, so the system always knows which capabilities are available, and at which version.
The catalog is grounded in established professional frameworks — Scrum, SAFe, ISTQB, TOGAF — which give structured methods and terminology to the capabilities it represents; deliverables can flow through MCP integrations (Jira, Confluence).
Persona disclaimer
Each agent defines a simulated expert persona used for prompting purposes. The certifications named in the agent identity blocks (PMP, CBAP, ISTQB, TOGAF, …) describe the persona's knowledge frame — they are not claims that the author, the software, or the model actually holds these certifications. Agent outputs are not professional, legal, financial, or medical advice.
Expertise that can actually be used
The catalog is not just a list of AI roles. Each agent maps a business request → the skills that answer it → the capability delivered, and the same automated run that rebuilds the index checks, on every change, that each of the 425 skills is actually reachable from the agents meant to use it — none left stranded, and no request pointing at a skill that is not there. This closes a common failure of AI systems: capabilities that exist on paper but cannot actually be reached or executed.
How to use the catalog
Start with the business outcome, not the agent. You don't browse the catalog to find interesting agents — you use it to produce a business deliverable. Give an agent your business context, and it takes on the corresponding professional role. No installation, no framework to learn.
From business need to deliverable
Business need → Expertise → Process → Controlled execution → Deliverable
For example — "We want to build an AI product, but the scope is still unclear." The system turns that request into a structured product-scoping process, pulls in the right expertise, and produces concrete deliverables along the way — instead of guessing at the scope that is missing.
How it fits together
Under the hood, the catalog is structured as Agent → Skills → Workflow → Runtime:
- An agent is a specialized professional role.
- It loads its own skills — the actionable sheets that define how that role performs tasks and produces deliverables.
- Agents are assembled into a workflow: a repeatable business process aimed at a defined outcome.
- The Agentic Runtime executes that process under controlled conditions, checking each step and recording what happened.
The architecture is technical; the outcome is business.
What are you trying to accomplish?
| You want to… | Use | Example |
|---|---|---|
| Produce a specific deliverable — a user story, analysis, wireframe or audit | a single agent | AGENT-PO-SAFE.md |
| Solve a problem spanning several areas of expertise | the orchestrator | AGENT-ORCHESTRATEUR-WORKFLOW.md |
| Run a repeatable business process end to end | the orchestrator + a workflow | WF-001 — AI Product Scoping |
One agent, one task — load the role, then give it your context, constraints and expected deliverable:
Read the file AGENT-PO-SAFE.md and take on this role.
Confirm you are ready by listing the available skills.
Then give it the specifics — for example:
"B2B SaaS onboarding, using SAFe terminology and practices, delivered over one Program Increment. Build a structured, prioritized backlog: decompose the scope into Epics, Features and INVEST user stories, prioritize the Features with WSJF, and give each user story acceptance criteria against the team's Definition of Done."
The agent turns the request into a structured, prioritized backlog, decomposing higher-level initiatives into implementable team-level work:
- Epic — Self-serve onboarding for B2B SaaS customers (large initiative,
prioritized at portfolio level with WSJF).
- Features (WSJF-ordered) — Guided workspace setup · Team invitation &
roles · CRM integration in a guided flow.
- User stories (INVEST) — "As an admin, I want to create a workspace in under two minutes, so that my team can start immediately" · "As a user, I want to connect our CRM in a guided flow, so that our data syncs automatically."
- Features (WSJF-ordered) — Guided workspace setup · Team invitation &
roles · CRM integration in a guided flow.
A full business process — hand the orchestrator a workflow to run:
Read the file AGENT-ORCHESTRATEUR-WORKFLOW.md and take on this role.
Then load workflows/WF-001-cadrage-produit-ia.md and start the workflow.
These are the catalog's own commands, taken verbatim from its START.md.
Deliverables can then flow out through the catalog's 3 MCP servers — for
example, a backlog pushed to Jira, a report published to Confluence, or a run
recorded in a mission log.
Browse the agents below · Open the catalog on GitHub →
Five areas of expertise
The catalog brings specialized capabilities together across five complementary domains. These are not isolated chat personas: each is a building block designed to be assembled into repeatable business processes. Each agent pairs a role with a focused skill folder it loads on demand.
Select the expertise you need. Assemble the process. Produce the deliverable.
Development & Engineering (16)
| Role | Focus |
|---|---|
| AI Python Developer | Builds LLM pipelines and agents in Python — LangChain/LangGraph, RAG (chunking, embedding, reranking), Hugging Face, PyTorch, fine-tuning (LoRA/QLoRA) and vector databases. |
| AI TypeScript Developer | Ships AI web apps in TypeScript — Next.js, Vercel AI SDK, MCP servers, streaming and Edge Functions. |
| Drupal / PHP Developer | Develops Drupal 10 / PHP — custom modules, Commerce 2.x, Twig templating, REST APIs and PHPUnit testing. |
| CMS Expert & Digital Platform Manager | Runs enterprise content platforms — AEM, Drupal 10, headless CMS, migrations, editorial governance, PIM/DAM, accessibility (RGAA) and SEO. |
| Tech Lead / AI Lead Developer | Owns technical quality — code review, application architecture, API design, CI/CD, OWASP security, technical-debt control and mentoring. |
| Business Intelligence Analyst | Turns data into decisions — dimensional modeling, Power BI/Fabric, Tableau, Looker, analytical SQL, KPIs and self-service BI. |
| PIM & Product Catalog Expert | Governs product information — catalog modeling, enrichment, channel syndication, ERP→PIM ETL, quality scoring, localization and generative AI for PIM. |
| DAM & Digital Assets Expert | Manages digital assets end to end — taxonomy, rights & licensing, renditions, multichannel distribution, brand portal, governance and visual AI. |
| AI Architect | Designs AI systems — RAG architectures, multi-agent orchestration, MCP/A2A, LLM evaluation and AI stack selection. |
| Solutions Architect (Enterprise) | Shapes enterprise IT — TOGAF ADM, ArchiMate 3, BDAT architecture, integration patterns, cloud migration and transformation roadmaps. |
| MLOps Engineer | Operationalizes ML — MLflow, Kubeflow, DVC, Feature Store, drift monitoring and ML CI/CD. |
| Data Engineer | Builds data pipelines — ETL/ELT, Spark, dbt, Airflow, Kafka, Databricks and lakehouse architectures. |
| Data Scientist | Models and predicts — supervised/unsupervised ML, statistics, experimentation, MLflow, Python and SQL. |
| DevOps / Cloud Engineer | Automates delivery and infrastructure — GitHub Actions CI/CD, Kubernetes, Terraform and Docker across AWS, GCP and Azure. |
| AI Security Expert | Secures AI systems — OWASP LLM Top 10, red teaming, CISSP practices and Zero Trust. |
| Prompt Engineer | Engineers and evaluates prompts — system-prompt design, few-shot, CoT, RAG prompts, multimodal and cost optimization. |
Agile, Product & Quality (11)
| Role | Focus |
|---|---|
| Product Owner Scrum | Owns the product backlog — user stories, prioritization (MoSCoW/RICE/Kano), story mapping, DoR/DoD and Scrum events; measures value with EBM. |
| Product Owner SAFe | Drives the program backlog — PI Planning, ART, WSJF prioritization, features and portfolio epics. |
| Scrum Master | Serves the team — event facilitation, coaching, agile metrics and scaling. |
| QA Agile (ISTQB) | Quality in-sprint — BDD/Gherkin, ATDD, exploratory testing and CI/CD integration. |
| QA Waterfall-Model (ISTQB) | Structured V-model testing — master test plan, acceptance, UAT and performance. |
| UX / UI Designer | Designs usable products — user research, personas, wireframing, Figma, design systems, WCAG and A/B testing. |
| Business Analyst / MOA | Bridges business and IT — elicitation, BPMN/UML, functional specifications, business UAT and change management (Agile + V-model). |
| Product Manager SAFe | Sets program direction — product vision, ART roadmap, customer-centricity, Lean business case and release strategy. |
| Release Train Engineer | Runs the ART — PI Planning facilitation, cross-team coordination, impediment removal, flow metrics and Inspect & Adapt. |
| Change Manager | Drives adoption — ADKAR, impact analysis, communication planning, resistance management and adoption measurement. |
| AI Methodology Auditor (Control & Challenge) | Independent control gate — Scrum/SAFe/ISTQB/PMI compliance audit, reasoning challenge, deliverable validation and catalog quality audit. |
Orchestration & Workflows (1)
| Role | Focus |
|---|---|
| Agentic Workflow Orchestrator | Runs multi-agent workflows — BPMN design, agent routing, context handoff, parallel orchestration, error recovery, output validation and monitoring. |
Management, Consulting & Content (9)
| Role | Focus |
|---|---|
| AI Project Manager | Delivers AI projects — scoping, planning, risk management, EVM and steering committees. |
| AI Consultant | Advises on AI adoption — maturity diagnostic, roadmap, ROI, benchmarking and PoC. |
| CDO / Chief AI Officer | Leads data & AI — data strategy, DAMA governance, OKR, data mesh and AI talent. |
| AI Trainer | Upskills teams — needs analysis, curriculum design, facilitation, e-learning and data literacy. |
| AI Growth / Marketing | Grows the product — acquisition, marketing automation, AI SEO, analytics and A/B testing. |
| AI Content Writer | Produces content at scale — copywriting, SEO, editorial prompting and multi-format output. |
| AI Legal | Keeps AI compliant — GDPR, AI Act, contracts and algorithmic accountability. |
| Financial Analyst | Builds the numbers — AI business case, transformation ROI, TCO/EVM, budgeting and executive-committee reporting. |
| Strategic Intelligence | Watches the field — AI/LLM and competitive intelligence, weak signals, tool benchmarking and periodic briefings. |
HR & Talent (1)
| Role | Focus |
|---|---|
| AI HR & IT Talent Acquisition | Sources and secures talent — IT/AI sourcing, workforce planning (GEPP), people analytics, ATS scoring and CV/deepfake fraud prevention. |
Skills (37 folders)
Each agent loads a matching skill folder holding its actionable content,
deliverables and output formats — the capabilities summarized in the Focus
column above. One folder per agent, except the two QA agents, which share
qa_testing/ (38 agents → 37 folders).
Development & Engineering (16) — dev_python_ia/ · dev_typescript_ia/ ·
dev_drupal/ · cms_digital/ · tech_lead/ · bi_analyst/ · pim_expert/ ·
dam_expert/ · ai_architect/ · solutions_architect/ · mlops_engineer/ ·
data_engineer/ · data_scientist/ · devops_cloud/ · securite_ia/ ·
prompt_engineer/
Agile, Product & Quality (10) — scrum/ · safe/ · scrum_master/ ·
qa_testing/ · ux_design/ · business_analyst/ · product_manager_safe/ ·
release_train_engineer/ · change_manager/ · critique_conformite/
Orchestration & Workflows (1) — orchestrateur_workflow/
Management, Consulting & Content (9) — chef_projet_ia/ · consultant_ia/ ·
cdo_directeur_ia/ · formateur_ia/ · growth_ia/ · redacteur_ia/ ·
juridique_ia/ · financial_analyst/ · veille_strategique/
HR & Talent (1) — rh_ia/
Why the catalog matters
A capable AI system needs more than a powerful model. It needs a defined set of capabilities that are:
- Specialized — each has a clear role and a bounded scope.
- Reusable — the same expertise supports several workflows.
- Composable — capabilities combine inside one process.
- Versioned — changes are tracked and reproduced.
- Validated — the reachability check above runs automatically on every change.
Together these turn AI expertise into an operational asset rather than a collection of prompts.
Where the expertise is used
-
Workflows
The catalog's capabilities are assembled into 10 structured business workflows — all ten have now run live, nine completed end to end and the tenth returned by its own quality gate.
-
Agentic Runtime
The Agentic Runtime executes these capabilities under controlled conditions — eval-gates enforced at every step, as a read-only consumer of pinned, versioned imports.
Catalog currently at v4.4.0 — versioned, with the full source in its
repository.