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>
24 lines
1.0 KiB
Markdown
24 lines
1.0 KiB
Markdown
# 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.
|