[alexsoyes]

🎯 What & why

Stacked on #1027. The plugin agents stay the single source and ship no model; this layer pins each role's model and effort at a chosen provider when the CLI installs them.

🛠️ How it works

• agent-models.json: role → tier (explorer, executor, validator light; reviewer standard) → model and effort per provider (openai, anthropic, mistral), with host-specific IDs where a host names a model differently.
• aidd translate … --provider <p> applies it through each host's agent function from feat(framework): ship generic sub-agents across six hosts #1027. A model an agent declares itself is kept; a provider the host cannot run is refused.
• make reload reads the provider per host from scripts/reload-agents.json; Claude now installs from its build so the model applies.

🧪 How to verify

cd cli && pnpm typecheck && pnpm test
node cli/dist/cli.js translate . --to opencode --as flat --out /tmp/x --provider openai

Observed: OpenCode loads the executor as {id: gpt-6-luna, providerID: openai, variant: medium}; a Codex session spawned the explorer on gpt-6-luna under a gpt-6.1-sol parent.

⚠️ Heads-up

Still to build in this PR, before ready:

• aidd setup --ide <host> [--provider <p>] [--main] [--global]: install in the current project by default, --global for the user's host directories; --ai and --scope stay as hidden aliases.
• --main: pin the main session's model, keeping the previous value.
• aidd clean: remove pinned models and restore the main one.
• make reload reduced to aidd setup --global ….

✅ I certify

• I DO CERTIFY I READ EACH LINE OF THE PULL REQUEST BECAUSE I AM A SOFTWARE ENGINEER, NOT A AI PUPPY.

[waewoo]

Thanks for this work!

I have a question about the hardcoded provider/model mapping in agent-models.json.

The current configuration defines a fixed set of providers and models. However, tools like Kilo Code and OpenCode support a much wider range of providers and models, including locally hosted models, custom API endpoints, and OpenAI-compatible providers.

While omitting --provider preserves model inheritance, it doesn't allow users to benefit from role-based model selection with their preferred models.

Would it make sense to support user-defined model mappings, allowing users to override the default models for each role (explorer, executor, validator, reviewer), regardless of the provider or model they use?

This could preserve the built-in defaults while keeping concrete model selection under user control, consistent with the discussion in #677.

Is this flexibility already planned, either within this PR or as a follow-up?