Files
cp1-agents/individual-agents/ALZ-F02: Assessment & Disposition Workflow Agent/README.md
Jonathan Boniface e22d037ab0 ALZ-F02 Assessment: individual agents, intake, rate card input, critic fixes
- Six declarative agents generated from the v1.1 stories (scaffold_from_stories.py, ADR 0001),
  and a 9-step workflow export (build_workflow.py).
- Code-only intake (ADR 0008) verifies the one Discovery package in code before any agent.
- The rate card is the run input rate_card_document; no upload step (ADR 0009).
- Critic 0.1.6 after live runs on crucible-kab-dev:
  - K-12 model independence checked at build time against each agent's deployed model
    (ADR 0010); the critic reports it from its two inputs;
  - receives in_cloud_estate for K-10; an unverified K-10 never fails the run alone;
  - K-08 allows the scoring note to list findings as advisory, as ALZ-ASM-04 must;
  - carries ALZ-ASM-03's and ALZ-ASM-04's rule sections, copied by the scaffold, so it can
    recalculate risk, confidence and TCO.
- The scaffold refuses to regenerate unless every story file is present.
- Case azure-crucible-kab-dev-assessment with an illustrative rate card.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-05 22:13:25 +00:00

88 lines
4.8 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# ALZ-F02 Assessment & Disposition: individual agents
Six declarative KAB agents, one per ALZ-ASM story, chained by one KAB workflow, behind a code-only intake step that
checks the Discovery package first (ADR 0008). They take an approved Discovery run's package and a rate card. For every application they produce a readiness profile, a migration risk score,
advisory security findings (in-cloud estates only), a 7R disposition with an indicative TCO, and an independent
critic's verdict. The Engagement Lead approves the verdict as `disposition_signoff`.
Requirements: ADO Feature 3214442, stories 3214459–3214464. The sub-agent pattern for this Feature isn't built yet,
and the two are kept separate.
## The workflow
`workflows/cp1-alz-assessment-individual.json`, built by `scripts/build_workflow.py`:
| Step | What | Inputs from |
|---|---|---|
| `a1` | Artifacts: the Discovery run's package, by exact id | The run (`artifact_ids`) |
| `p1` | Parse Assessment Files over `a1`'s records | `a1.artifacts` |
| `s0` | [Assessment Intake](cp1-assessment-intake-agent/) (ADR 0008): checks the package in code (one package, every part's sha256, complete, critic pass, the named Discovery run); a refused package fails the run here | `p1` |
| `s1` | [Assessment Planner](cp1-assessment-planner-agent/) (ALZ-ASM-01): checks the preconditions and plans, or refuses | `p1`, the run's `rate_card_document`, after `s0` |
| `s2` | [Cloud Readiness Profiler](cp1-cloud-readiness-profiler-agent/) (ALZ-ASM-02) | `p1`, `s1` |
| `s3` | [Migration Risk Scorer](cp1-migration-risk-scorer-agent/) (ALZ-ASM-03) | `p1`, `s1` |
| `s4` | [Security Posture Assessment](cp1-security-posture-assessment-agent/) (ALZ-ASM-05): not applicable on-premises | `p1`, `s1` |
| `s5` | [7R Disposition & TCO](cp1-disposition-tco-agent/) (ALZ-ASM-04) | `p1`, the run's `rate_card_document`, `s1`, `s2`, `s3`, `s4` |
| `s6` | [Assessment Critic & Sign-off Router](cp1-assessment-critic-agent/) (ALZ-ASM-06): **approval step** | All of the above |
`s2`, `s3` and `s4` run in parallel.
## Starting a run
1. **Choose a Discovery run** whose critic verdict is `pass` and which was approved. Note its execution id and the
artifact id of its **package** step output (F01 ADR 0020). KAB names a run's step artifacts by position, not step
id: the package is `{execution_id}_step_12` in the current Discovery workflow. Select only that one artifact: the
intake refuses anything else.
2. **Start the workflow** with these values:
- `engagement_id`
- `run_id`
- `discovery_execution_id`
- `discovery_approval_ref`: the decision record approving the Discovery run
- `artifact_ids`: the Discovery package artifact
- `rate_card_document`: the rate card's JSON, as text (ADR 0009). There's no upload: an uploaded file would reach
the parse step and be read in place of the Discovery package
Alternatively, fix the artifact ids in the export with
`python3 scripts/build_workflow.py --artifact-id <id> ...`.
3. **Expect the intake to fail the run** if the package is missing, changed, incomplete, from a failed critic, or
from another Discovery run or engagement: its error lists every reason.
4. **Expect a refusal, not a plan,** if no approval is given or there's no rate card. The critic then fails the run.
## Rate card
The rate card is a JSON file with `rate_card_id`, `currency` and `items`. Each item has:
- `rate_id`
- `category`
- `description`
- `annual_rate`
- `equivalent_for`: the services, products or asset classes it can replace
- `licence_terms`
An illustrative sample for the Meridian Retail Group estate is in
[workflows/cases/meridian-assessment/rate-card.json](workflows/cases/meridian-assessment/rate-card.json). TCO is
indicative; the Business Case (ALZ-BCS-02) gives the authoritative figure.
## MCP servers
**None added.** Only ALZ-ASM-02 and ALZ-ASM-05 call tools, all read-only and all on existing servers in
[`mcp-servers/`](../../mcp-servers/): `azure-discovery`, `aws-discovery` and `gcp-discovery`, including their
Well-Architected guidance tools. All three must be registered and healthy before those two agents can start (ADR 0007). The other four agents call no tool.
## Decisions
The [decision records](docs/adr/README.md) cover how this pattern meets or interprets the stories:
- Scoring as prompt rules that the critic recalculates.
- Discovery's outputs arriving as parsed text, with the run id and approval as run inputs.
- The rate card format.
- Sign-off as the critic step's approval.
- The models used.
- The tool-call check this pattern can't make.
- The MCP servers that must be registered first.
## Status
Built and checked offline: six folders pass `check_agents.py`, the workflow export builds, and all 25 declared tools
exist on their servers. Nothing is deployed, and no eval case or workflow run has been executed yet. The remaining
steps are in [.agents/PLAN.md](.agents/PLAN.md).