docs/releases/0.6.0.md

# codex-sdlc 0.6.0

Save a project-wide model preset for every new SDLC run with `/sdlc --save-my-token`, and return to inherited models with `/sdlc --normal`.

## Changes

- Add the `sdlc` entry skill for project mode changes and routing feature requests to PM.
- Token-saving mode inherits PM, uses `gpt-6-sol` with `high` reasoning for BA, and `gpt-6-luna` with `xhigh` reasoning for backend, web/mobile frontend, and QC.
- Normal mode removes model, effort, and fallback overrides for those five roles. It restores inheritance rather than earlier custom settings.
- Store settings in the coordinator's `.sdlc/project.yaml` for every future run. Preserve existing run snapshots, other projects, and optional AI Product Owner settings.
- Add `configure-agents --save-my-token` and `configure-agents --normal`, including dry-run previews and conflict rejection. No schema migration is required.

The slash forms are skill-handled chat shorthand, not registered native slash commands. If a client rejects them, use `$sdlc --save-my-token` or `$sdlc --normal`. Host model availability is checked at dispatch; the preset adds no implicit fallback or guarantee of token savings.

## Upgrade

Update the plugin to 0.6.0, then upgrade the coordinator's repository runtime separately:

```sh
npx --yes codex-sdlc@0.6.0 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. Apply a mode before starting a new run:

```sh
node .sdlc/runtime.cjs configure-agents --save-my-token
# Return future runs to inherited delivery-role models:
node .sdlc/runtime.cjs configure-agents --normal
```

See [project model modes](https://github.com/prime-vimihi/codex-sdlc/blob/v0.6.0/docs/agent-models.md#project-model-modes) for role settings and behavior.

## Validation

All 51 automated tests, TypeScript checks, the build, and skill validation passed locally. An independent built-CLI check exercised both previews, exact saved settings, future run snapshots, mobile routing, normal reset, existing-run preservation, and invalid targets or conflicting flags. The npm and local marketplace bundles include the new entry skill.

Actual model dispatch and chat shorthand execution were not exercised for this change. Native Linux and Windows qualification and a complete delivery across multiple models remain deferred.