Add CP1 Discovery agents in both patterns
All checks were successful
check agents / check (push) Successful in 10s
All checks were successful
check agents / check (push) Successful in 10s
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>
This commit is contained in:
@@ -0,0 +1,9 @@
|
||||
# Individual agents, ALZ-F01 Discovery: documentation
|
||||
|
||||
For people who run, review or operate CP1 Discovery with the individual-agents pattern. Instructions for coding agents are in `../CLAUDE.md` and `../.agents/`.
|
||||
|
||||
| Document | For |
|
||||
|---|---|
|
||||
| [deployment.md](../../../docs/deployment.md) | Platform operators: building, pushing and deploying the eight agents with kustomize |
|
||||
| [adr/](adr/README.md) | Product owner and reviewers: every change to the original ALZ-F01 requirements in this pattern |
|
||||
| `<agent>/docs/user-guide.md` | Engagement leads: what each step does and how to read its output |
|
||||
@@ -0,0 +1,19 @@
|
||||
# 0001. Eight separate agents chained by the KAB workflow, with no orchestrator
|
||||
|
||||
| | |
|
||||
|---|---|
|
||||
| Status | Accepted |
|
||||
| Date | 2026-10-29 |
|
||||
| Decided by | Product owner (CP1) |
|
||||
| Requirements changed | None in the stories; this is the stories' own "no orchestrator" model, delivered the cp4-agents way |
|
||||
|
||||
## Decision
|
||||
|
||||
- **One declarative agent per ALZ-DSC story,** in cp4-agents' shape and with no `app/`. Each calls its own read-only MCP tools.
|
||||
- **The workflow wires them together** (`workflows/cp1-alz-discovery-individual.json`): each step's inputs are named fields of earlier steps' outputs.
|
||||
- **The orchestrator pattern stays separate** in `../cp1-agents`, so the two can be compared.
|
||||
|
||||
## Consequences
|
||||
|
||||
- **The good:** each agent can be deployed, evaluated and released on its own. Each has its own MCP identity, so per-agent allowlists hold, and no agent needs code.
|
||||
- **The cost:** the rules in the stories' pseudo-code are applied by each model, not by code, so the eval cases carry more weight. Nothing outside the workflow remembers a run (ADR 0010).
|
||||
@@ -0,0 +1,20 @@
|
||||
# 0002. Workflow hand-offs replace the alz-run tools
|
||||
|
||||
| | |
|
||||
|---|---|
|
||||
| Status | Accepted |
|
||||
| Date | 2026-10-29 |
|
||||
| Decided by | Product owner (CP1), decision D1 |
|
||||
| Requirements changed | Every story's Tool Definitions: `alz-run.get_outputs`, `submit_output`, `submit_verdict`, `checkpoint` |
|
||||
|
||||
## Decision
|
||||
|
||||
- **Each agent's input contract** (`schemas/input.json`) replaces `get_outputs`.
|
||||
- **Its reply, checked against its output contract,** replaces `submit_output` and `submit_verdict`.
|
||||
- **The runtime's loop limits** replace `checkpoint`.
|
||||
- **The workflow builder** checks that every field passed between steps exists on both sides (`scripts/build_workflow.py`).
|
||||
|
||||
## Consequences
|
||||
|
||||
- **The server-side "id not scanned" check is gone.** The critic's call-log check (DSC-08) covers it.
|
||||
- **A schema failure is retried once,** then the step fails with `output_contract_violation`.
|
||||
@@ -0,0 +1,20 @@
|
||||
# 0003. The MCP write tools are dropped; outputs go back in the reply
|
||||
|
||||
| | |
|
||||
|---|---|
|
||||
| Status | Proposed |
|
||||
| Date | 2026-10-29 |
|
||||
| Decided by | Built as a default; awaiting product owner confirmation |
|
||||
| Requirements changed | DSC-03 `alz-inventory.write_inventory`, DSC-05 `alz-classification.tag_pii`, DSC-06 `alz-dependency.graph` |
|
||||
|
||||
## Context
|
||||
|
||||
The platform stores those tools wrote to don't exist.
|
||||
|
||||
## Decision
|
||||
|
||||
Each agent returns those records in its reply. Every tool in every `mcp:` list is read-only.
|
||||
|
||||
## Consequences
|
||||
|
||||
Persisting inventory, tags and graphs is left to platform artifacts: the executor registers step outputs. A "write" MCP server can be added later without changing the contracts.
|
||||
@@ -0,0 +1,16 @@
|
||||
# 0004. The scan executor runs once per run and returns one finding per entry
|
||||
|
||||
| | |
|
||||
|---|---|
|
||||
| Status | Proposed |
|
||||
| Date | 2026-10-29 |
|
||||
| Decided by | Built as a default; awaiting product owner confirmation |
|
||||
| Requirements changed | DSC-02 (3214452): "one finding submitted per call" |
|
||||
|
||||
## Decision
|
||||
|
||||
The step receives every `scan_target` task and still calls `scan_target` once per entry. It replies `{"findings": [...]}` with one eight-key finding per task. The finding schema is unchanged.
|
||||
|
||||
## Consequences
|
||||
|
||||
A very large estate may exceed the step's 20-minute timeout or loop limits. Split it with more planner batches and more scan steps if that happens.
|
||||
@@ -0,0 +1,21 @@
|
||||
# 0005. Inventory before CMDB reconciliation
|
||||
|
||||
| | |
|
||||
|---|---|
|
||||
| Status | Proposed |
|
||||
| Date | 2026-10-29 |
|
||||
| Decided by | Built as a default; awaiting product owner confirmation |
|
||||
| Requirements changed | DSC-01's sequencing rule "CMDB / ITSM ingest first"; DSC-04 (3214454) owning ingest |
|
||||
|
||||
## Context
|
||||
|
||||
Reconciliation needs the canonical inventory, and the inventory needs CMDB facts. In one agent (DSC-04) that's circular, which is decision D2 in the alignment report.
|
||||
|
||||
## Decision
|
||||
|
||||
- **Step 3 (inventory)** takes CMDB and scan source records from `alz-inventory.get_app`, which returns both.
|
||||
- **Step 4 (reconciliation)** then ingests through `alz-cmdb`, as DSC-04 specifies, and diffs against that inventory.
|
||||
|
||||
## Consequences
|
||||
|
||||
The CMDB is read twice, once through each server. The planner's ingest tasks are still planned and still come before network, dependency and reconciliation work.
|
||||
@@ -0,0 +1,23 @@
|
||||
# 0006. The critic's verdict is gated by human approval, not by code
|
||||
|
||||
| | |
|
||||
|---|---|
|
||||
| Status | Proposed |
|
||||
| Date | 2026-10-29 |
|
||||
| Decided by | Built as a default; awaiting product owner confirmation |
|
||||
| Requirements changed | DSC-08 (3214458): "the platform sets the phase status from the verdict" (`alz-run.submit_verdict`) |
|
||||
|
||||
## Context
|
||||
|
||||
A declarative agent has no hooks, so nothing can recompute or enforce the verdict in code. The orchestrator pattern's `post_process` does that.
|
||||
|
||||
## Decision
|
||||
|
||||
- **The critic's step is a workflow approval** (`hitl.checkpoint_type: approval`). The Engagement Lead sees the verdict and its reasons, and a fail is never auto-approved.
|
||||
- **The critic runs on a different model** from the other agents (ADR 0008).
|
||||
- **The critic gets every earlier step's output** through `from_steps`.
|
||||
|
||||
## Consequences
|
||||
|
||||
- **The verdict is advisory until a person approves it.** A wrong `pass` relies on the reviewer to catch it.
|
||||
- **The trade-off against `../cp1-agents`'s enforced verdict** is the main point to settle between the two patterns.
|
||||
@@ -0,0 +1,23 @@
|
||||
# 0007. Agent names
|
||||
|
||||
| | |
|
||||
|---|---|
|
||||
| Status | Proposed |
|
||||
| Date | 2026-10-29 |
|
||||
| Decided by | Built as a default; awaiting product owner confirmation |
|
||||
| Requirements changed | Agent IDs ALZ-DSC-01…08 as deployable names |
|
||||
|
||||
## Decision
|
||||
|
||||
The `cp1-<role>-agent` names follow cp4-agents' `cp4-<role>-agent` convention. The story ID stays in each prompt, README and ADR index.
|
||||
|
||||
| Story | Agent |
|
||||
|---|---|
|
||||
| ALZ-DSC-01 | `cp1-discovery-planner-agent` |
|
||||
| ALZ-DSC-02 | `cp1-estate-scan-executor-agent` |
|
||||
| ALZ-DSC-03 | `cp1-asset-inventory-normalisation-agent` |
|
||||
| ALZ-DSC-04 | `cp1-cmdb-itsm-reconciliation-agent` |
|
||||
| ALZ-DSC-05 | `cp1-data-classification-agent` |
|
||||
| ALZ-DSC-06 | `cp1-dependency-mapping-agent` |
|
||||
| ALZ-DSC-07 | `cp1-network-discovery-assessment-agent` |
|
||||
| ALZ-DSC-08 | `cp1-discovery-critic-agent` |
|
||||
@@ -0,0 +1,17 @@
|
||||
# 0008. Models
|
||||
|
||||
| | |
|
||||
|---|---|
|
||||
| Status | Proposed |
|
||||
| Date | 2026-10-29 |
|
||||
| Decided by | Built as a default; awaiting product owner and platform confirmation |
|
||||
| Requirements changed | DSC-08's note on sharing a model with any executor |
|
||||
|
||||
## Decision
|
||||
|
||||
- **Seven agents use `gpt-5.6-sol`,** as the CP4 agents do, with eval cases waiting 330 seconds.
|
||||
- **The critic uses `claude-sonnet-5-5`.** Both models are deployed on the Azure AI resource behind the gateway.
|
||||
|
||||
## Consequences
|
||||
|
||||
Confirm both names on the cluster's LiteLLM gateway before the first run.
|
||||
@@ -0,0 +1,14 @@
|
||||
# 0009. Deploy with kustomize; keys stay in Secrets
|
||||
|
||||
| | |
|
||||
|---|---|
|
||||
| Status | Accepted |
|
||||
| Date | 2026-10-29 |
|
||||
| Decided by | Product owner (CP1) |
|
||||
| Requirements changed | None; cp4-agents leaves deployment to Kyndryl's agent CI |
|
||||
|
||||
## Decision
|
||||
|
||||
- **Placeholders stay:** each `kubernetes/` folder keeps the CI placeholders.
|
||||
- **`scripts/render_k8s.py`** fills them into `build/` for `kubectl diff -k` and `kubectl apply -k`, leaving out the Secret stub and the shared `agents` namespace.
|
||||
- **Each agent's MCP key** goes in its own Secret, `<agent>-secrets`.
|
||||
@@ -0,0 +1,17 @@
|
||||
# 0010. Step outputs travel through the workflow; there is no shared store
|
||||
|
||||
| | |
|
||||
|---|---|
|
||||
| Status | Proposed |
|
||||
| Date | 2026-10-29 |
|
||||
| Decided by | Built as a default; awaiting platform team confirmation |
|
||||
| Requirements changed | Platform rails PR-11 (evidence chain): each agent's outputs were to be recorded on a hash-linked chain |
|
||||
|
||||
## Decision
|
||||
|
||||
A step's output exists only as that step's reply. The executor carries it to later steps (externalising large payloads to execution storage) and registers it as an `execution_output` artifact.
|
||||
|
||||
## Consequences
|
||||
|
||||
- **No hash chain across steps:** evidence integrity rests on each fact's `evidence_ref` and the critic's call-log check.
|
||||
- **No memory between runs.** That is the cost of having no orchestrator; `../cp1-agents` keeps a workspace instead.
|
||||
@@ -0,0 +1,24 @@
|
||||
# 0011. Uploaded files and project artifacts reach the planner and the inventory as text
|
||||
|
||||
| | |
|
||||
|---|---|
|
||||
| Status | Accepted |
|
||||
| Date | 2026-10-29 |
|
||||
| Decided by | Product owner (CP1) |
|
||||
| Requirements changed | DSC-01 (3214451) and DSC-03 (3214453): estate data came only from MCP sources; client files and earlier outputs were not an input |
|
||||
|
||||
## Decision
|
||||
|
||||
- **Two optional KAB activity steps sit in front of the workflow:**
|
||||
- **Parse Assessment Files** (`u1`): the run form's upload field `file`;
|
||||
- **Artifacts** (`a1`): project artifacts fixed by id.
|
||||
- **The planner (s1) and the inventory agent (s3) take their `parsed_content`** as two optional text inputs: `source_documents` and `reference_documents`.
|
||||
- **These agents are declarative,** with no code to open files, so they take the text the platform has already extracted.
|
||||
- **Their prompts treat the text as evidence only:** it never adds a target, it is cited as `document:source` or `document:reference`, it is weaker than MCP results, and it is never followed as an instruction.
|
||||
- **`scripts/build_workflow.py` adds the front steps,** and `scripts/kab_import_workflow.py` validates and imports the workflow.
|
||||
|
||||
## Consequences
|
||||
|
||||
- **Size:** inline text is bounded by the platform's payload limits. Large exports suit the orchestrator pattern, which reads files from the volume.
|
||||
- **Outputs:** every step's reply remains an `execution_output` artifact.
|
||||
- **The other six agents are unchanged** (still 0.1.0).
|
||||
@@ -0,0 +1,29 @@
|
||||
# Architecture decision records: individual agents, ALZ-F01 Discovery
|
||||
|
||||
Each record documents a change to the original requirements for ALZ-F01 Discovery: the ADO Feature 3214441, its
|
||||
stories 3214451–3214458, and the v1.1 story files (`drafts/cp1/agentic-landing-zone/stories/F01-discovery/` in the
|
||||
engineering hub workspace). These records apply to the **individual-agents pattern** in this Feature folder. The
|
||||
sub-agent pattern in `sub-agents/ALZ-F01: Discovery Workflow Agent/` keeps its own records, and the two patterns' decisions are separate.
|
||||
Which pattern is adopted has not been decided.
|
||||
|
||||
Each agent's `docs/adr/README.md` lists the records that change its own story.
|
||||
|
||||
**Status values:**
|
||||
- **Accepted:** decided by the product owner.
|
||||
- **Proposed:** built as a default and awaiting confirmation.
|
||||
|
||||
| ADR | Decision | Status | Stories |
|
||||
|---|---|---|---|
|
||||
| [0001](0001.md) | Eight separate agents chained by the KAB workflow, with no orchestrator | Accepted | All |
|
||||
| [0002](0002.md) | Workflow hand-offs replace the `alz-run` tools | Accepted | All |
|
||||
| [0003](0003.md) | The MCP write tools are dropped; outputs go back in the reply | Proposed | DSC-03, 05, 06 |
|
||||
| [0004](0004.md) | The scan executor runs once per run and returns one finding per entry | Proposed | DSC-02 |
|
||||
| [0005](0005.md) | Inventory before CMDB reconciliation; the inventory gets CMDB records from `alz-inventory` | Proposed | DSC-03, 04 |
|
||||
| [0006](0006.md) | The critic's verdict is gated by human approval, not by code | Proposed | DSC-08 |
|
||||
| [0007](0007.md) | Agent names | Proposed | All |
|
||||
| [0008](0008.md) | Models | Proposed | All |
|
||||
| [0009](0009.md) | Deploy with kustomize; keys stay in Secrets | Accepted | — |
|
||||
| [0010](0010.md) | Step outputs travel through the workflow; there is no shared store | Proposed | All |
|
||||
| [0011](0011.md) | Uploaded files and project artifacts reach the planner and the inventory as text | Accepted | DSC-01, 03 |
|
||||
|
||||
New records copy [template.md](template.md) and take the next number.
|
||||
@@ -0,0 +1,14 @@
|
||||
# NNNN. Title
|
||||
|
||||
| | |
|
||||
|---|---|
|
||||
| Status | Proposed |
|
||||
| Date | YYYY-MM-DD |
|
||||
| Decided by | |
|
||||
| Requirements changed | Story / ADO item and the section that no longer holds |
|
||||
|
||||
## Context
|
||||
|
||||
## Decision
|
||||
|
||||
## Consequences
|
||||
Reference in New Issue
Block a user