ALZ-F01 Discovery as a sub-agent orchestrator (sub-agents/) and as eight individual declarative agents (individual-agents/), with their KAB workflow exports. Shared check, render and import scripts, the alz-discovery-sample MCP server, the crucible-kab-dev deploy profile, deployment docs, platform issue reports and the source material. Agent steps time out at 20 minutes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ALZ-F01 Discovery: sub-agent pattern
The ALZ-F01 Discovery Feature (ADO 3214441) as one orchestrator agent whose eight sub-agents are the Feature's stories. It runs in four phases (plan, collect, analyse, critique). The orchestrator keeps a hash-chained run workspace, and the critic's verdict is enforced in code.
This is one of two patterns for CP1 Discovery, and which one is adopted hasn't been decided. The other is individual agents: eight separate agents chained by the workflow.
| Agent | What it does |
|---|---|
cp1-alz-discovery-orchestrator-agent |
Every phase of discovery. It makes all the MCP calls itself and delegates to sub-agents dsc_01_planner … dsc_08_critic. |
These agents carry app/ for their sub-agents, functions and hooks, and the shared check allows it (ADR 0008). Each decision is an ADR in the agent's docs/adr/.
Layout
| Path | What it is |
|---|---|
cp1-alz-discovery-orchestrator-agent/ |
The agent. README.md and docs/ are for people; CLAUDE.md and .agents/ are for coding agents. |
workflows/ |
This Feature's KAB workflow export: four phase steps, plus an upload step at the front |
scripts/ |
build_workflow.py (writes workflows/) and dry_run.py (tests the agent's code with no model and no MCP server) |
These are shared with the other pattern and live at the repository root:
- the MCP server
alz-discovery-sample(../../mcp-servers/); - the check, render and import scripts (
../../scripts/); - the deploy profile and the deployment guide.
Files and artifacts in a run
The workflow can start with two KAB activity steps (ADR 0012):
- Parse Assessment Files: the run form's upload field.
- Artifacts: project artifacts fixed by id.
Only the plan step takes them. It imports them into the run workspace, and the sub-agents get records only.
To build the workflow and apply it to KAB, run from the repository root:
python3 scripts/kab_import_workflow.py "sub-agents/ALZ-F01: Discovery Workflow Agent" --project-id <id> # validate only
python3 scripts/kab_import_workflow.py "sub-agents/ALZ-F01: Discovery Workflow Agent" --project-id <id> --apply # import as a draft
The script reads your KAB access token from ~/.kab_token. Importing needs the Editor role.
Check it
From this folder:
python3 ../../scripts/check_agents.py --runtime-image cruciblecoredevacr.azurecr.io/kab/agent-runtime-adk:0.35.7 cp1-alz-discovery-orchestrator-agent
python3 scripts/dry_run.py cp1-alz-discovery-orchestrator-agent "$(mktemp -d)"
Running the agent locally is described in its README. Deploying it is in docs/deployment.md.