docs/releases/0.5.0.md

# codex-sdlc 0.5.0

Choose a model for each SDLC role and add an advisory AI Product Owner review while keeping final acceptance with the user.

## Changes

- Configure PM, BA, backend, frontend, QC, and AI Product Owner models during `init` or later with `configure-agents`. Optionally set reasoning effort and an explicit fallback.
- Snapshot role settings into each new run. Existing runs keep their settings; frontend settings cover both web and mobile, including applications in separate repositories.
- Resolve dispatch plans against the current host's model, reasoning, and inheritance capabilities. A configured role requires a recorded agent dispatch before it starts. Unavailable selections stop dispatch unless an explicitly configured fallback matches.
- Record requested and selected models, real returned agent IDs, and fallback use. Actual model/effort stays unknown unless the host reports it.
- Run configured PM work in task-only PM agents. The current conversation dispatches tasks and handles user communication without claiming to change its own model.
- Add the optional `sdlc-po` skill and `PO-001` review after QC. Its artifact must cover every authoritative acceptance criterion and cite existing run artifacts. It offers a recommendation; only a human Product Owner decision accepts delivery.
- Generate permissions without YAML aliases so the strict command-line authority publisher can load both new and upgraded installations.
- Include setup guidance and examples for configuring, resetting, and inspecting role models.

## Upgrade

Install the 0.5.0 plugin, then upgrade the coordinator's repository runtime separately:

```sh
npx --yes codex-sdlc@0.5.0 upgrade --root /absolute/path/to/coordinator --dry-run
```

Apply the same command without `--dry-run`, then run `node .sdlc/runtime.cjs restore`, `doctor`, and `validate-config` from the coordinator. Configure role models before starting a new run. Existing runs and single-/multi-repository schema families remain supported.

See [role model guidance](../agent-models.md) for examples. Model names, availability, and supported reasoning depend on the Codex host; this release does not grant model access or use separate API credentials.

## Validation

TypeScript checks, the automated test suite, plugin/skill validation, and a local PM dispatch smoke test cover this release. The live test requests Sol and exercises spawn, dispatch recording, activation, and PM intake handoff; host-unreported actual model metadata remains unknown. It is not a complete multi-model delivery test. Native Linux and Windows qualification remains deferred.