# gsd-hermes v1.2.0 — Cross-Provider Agent Execution

Draft GitHub release copy for the next downstream feature release.

## Summary

`gsd-hermes@1.2.0` is the release that turns Cross-Provider Agent Execution into a first-class downstream capability.

The core change is that Hermes becomes a truthful multi-provider runtime adapter for GSD: planner, researcher, checker, executor, and verifier no longer need to collapse onto one pretend runtime model. When a provider/model binding is configured explicitly, GSD now either honors it directly or fails fast with an actionable error instead of silently falling back.

## Highlights

- **Cross-Provider Agent Execution** — run different GSD agents on different providers/models under `runtime: "hermes"` while preserving the standard upstream workflow shape.
- **Strict runtime/model binding** — explicit provider/model bindings are enforced exactly or rejected with clear diagnostics; silent fallback is no longer accepted.
- **Explicit external fallback** — `cross_ai_execution` stays available, but only as a deliberate fallback path rather than an invisible provider-translation layer.
- **Clearer operator visibility** — init/workflow/doc surfaces now expose the four runtime-model paths consistently: explicit binding, `inherit`, runtime-default omission, and explicit external fallback.

## Why this release matters

Before this release, mixed-provider setups risked being flattened into misleading runtime assumptions or hidden fallback behavior. With v1.2.0:

- runtime and provider/model are treated as separate concepts
- Hermes direct bindings remain direct when supported
- unsupported direct bindings fail before plan/execute work starts
- external delegation requires explicit opt-in instead of happening silently

That makes `gsd-hermes` much safer for teams that want to mix LLM vendors intentionally without losing confidence in what model actually executed each step.

## Example configuration

```json
{
  "runtime": "hermes",
  "model_profile": "inherit",
  "resolve_model_ids": "omit",
  "model_overrides": {
    "gsd-phase-researcher": "claude-opus-4-7",
    "gsd-planner": "claude-opus-4-7",
    "gsd-plan-checker": "openai/gpt-5.4",
    "gsd-executor": "openai/gpt-5.4",
    "gsd-verifier": "openai/gpt-5.4"
  }
}
```

## Install

```bash
npx gsd-hermes@latest
```

For unreleased main-branch testing before npm publish:

```bash
npx --yes github:pingchesu/gsd-hermes
```

## Verification

Recommended release verification references:

- PR: https://github.com/pingchesu/gsd-hermes/pull/15
- Milestone tag: `v1.2`
- Canonical config docs: `docs/CONFIGURATION.md`
- Command behavior docs: `docs/COMMANDS.md`
- Publish procedure: `docs/npm-publishing.md`

## Suggested release title

`gsd-hermes v1.2.0 — Cross-Provider Agent Execution`