azure-discovery, aws-discovery and gcp-discovery collect read-only metadata on servers, workloads, Kubernetes clusters, networks, data stores, policies and provider recommendations. Credentials arrive per call from the vault. Replies are paged, scrubbed of secrets and anonymised, and throttled calls back off. cloud-knowledge searches each provider's Well-Architected guidance from fixed lists only, so no customer detail can leave the platform. Adds shared skills, the render and test tooling, a public Python Dockerfile until the hardened base image is reachable, the customer access guide, the implementation plan and the Vault Service platform issue. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
CP1 agents
The KAB agents and workflows for the CP1 Epic "Agentic Landing Zone" (ADO 3214440), built on the KAB agent runtime in KAB's agent shape (agent.yaml, prompt.md, schemas/, eval/).
Each Feature is built in two patterns, and which one is adopted hasn't been decided:
| Pattern | Folder | Shape |
|---|---|---|
| Sub-agents | sub-agents/ |
One orchestrator agent per Feature, with the Feature's stories as its sub-agents. It keeps a shared run workspace, and a verdict enforced in code. |
| Individual agents | individual-agents/ |
One declarative agent per story, chained by the KAB workflow, cp4-agents style |
| Feature | Sub-agents | Individual agents |
|---|---|---|
| ALZ-F01 Discovery (3214441) | built: 1 orchestrator, 8 sub-agents | built: 8 agents |
| ALZ-F02 Assessment & Disposition | not started | not started |
| ALZ-F03 Strategy Synthesis | not started | not started |
Layout
cp1-agents/
├── sub-agents/<Feature>/ the sub-agent pattern, one folder per Feature
│ ├── cp1-…-orchestrator-agent/ the agent (Dockerfile, kubernetes/, docs/, .agents/)
│ ├── workflows/ the Feature's KAB workflow export
│ └── scripts/ scripts only this Feature uses (build_workflow.py, dry_run.py)
├── individual-agents/<Feature>/ the individual-agents pattern, one folder per Feature
│ ├── cp1-…-agent/ … one agent per story
│ ├── docs/adr/ .agents/ the pattern's decisions and plan for this Feature
│ ├── workflows/ scripts/
├── mcp-servers/ MCP servers shared by both patterns (alz-discovery-sample)
├── scripts/ shared: check_agents, render_k8s, kab_import_workflow, kab_front_steps, MCP server tools
├── deploy/ deploy profiles: where to deploy, never keys
└── docs/
├── deployment.md building, rendering and deploying every component
├── platform-issues/ KAB platform problems found while building these agents
├── source-material/ the inputs: agent shapes, KAB workflow exports, workflow documentation
└── strategy-outline/
Each agent keeps its own Dockerfile and kubernetes/ folder. Kyndryl's agent CI builds one folder at a time, and the kubernetes/ files keep the CI's placeholders. Everything else that the two patterns or several Features share exists once, at the root.
The Feature folder names contain spaces and a colon. Quote them in a shell. Git can't check out a path that contains : on Windows, so clone the repository in WSL or on Linux or macOS.
Common tasks
From the repository root:
# Check every agent folder
find sub-agents individual-agents -mindepth 3 -maxdepth 3 -name agent.yaml -print0 | xargs -0 -n1 dirname -z \
| xargs -0 python3 scripts/check_agents.py --runtime-image cruciblecoredevacr.azurecr.io/kab/agent-runtime-adk:0.35.7
# Rebuild a Feature's workflow export, then validate it against KAB and import it
python3 "sub-agents/ALZ-F01: Discovery Workflow Agent/scripts/build_workflow.py"
python3 scripts/kab_import_workflow.py "sub-agents/ALZ-F01: Discovery Workflow Agent" --project-id <id> [--apply]
# Render the manifests for a deploy
python3 scripts/render_k8s.py deploy/crucible-kab-dev.yaml [<component>]
The secrets, the MCP gateway registration and the order of the steps are in docs/deployment.md.
Checks
.gitea/workflows/check.yml runs on every push:
scripts/check_agents.py: cp4-agents' folder rules (pinned runtime image, Kyndryl CI's placeholders, no secrets). An agent may carryapp/only if it declares functions, hooks or sub-agents.- The orchestrator's dry run: its code, with no model and no MCP server.
- Every Feature's workflow export must match its agents' contracts.