refactor: add per-agent types and register setup adapters

[Stephen Belanger (Qard)]

Summary

Based on main now that #136 has merged.

This adds the agent model for the rest of the reorganization:

• One concrete type per agent.
• One trait per capability, which the agent types implement.
• Each agent registers itself with the capabilities it supports.

A command looks agents up in its own registry, so an agent that lacks a capability simply isn't there. There is no central descriptor listing every agent's features.

// src/agents/mod.rs
pub(crate) struct Identity { id, source, aliases, display_name }
pub(crate) trait Agent: Send + Sync + 'static { fn identity(&self) -> &'static Identity; }
pub(crate) struct Agents<T: ?Sized> { … }        // get(name) by id, source or alias
pub(crate) struct Registrar { pub setup: Agents<dyn Setup> }
pub(crate) trait Register { fn register(self: Arc<Self>, registrar: &mut Registrar); }
pub(crate) fn registrar() -> &'static Registrar  // the one place every agent is listed

// src/agents/grok.rs
impl Register for Grok {
    fn register(self: Arc<Self>, registrar: &mut Registrar) {
        registrar.setup.insert(self);
    }
}

Setup is the first capability:

• Each setup/<agent>.rs now has impl Setup for <Agent> instead of free enable/disable/update/stale functions.
• run_enable, run_disable, run_update and update_warning dispatch through the registry instead of matching on SetupAgent.
• The doctor's per-agent activation warnings move out of trace_runtime.rs into the Grok and Cursor Setup impls (activation_warning).

Identities replace duplicated tables. The clap enums (SetupAgent, DoctorAgent, ImportSource) still define the CLI surface, so value names and aliases are unchanged. Each enum gains identity(), which maps a variant to its agent. That replaces the id and display-name tables duplicated in:

• setup (agent_details)
• DoctorAgent::source/display_name
• transcript_import::source_name

A new test checks that every SetupAgent variant has a registered setup adapter.

The next PR registers the remaining capabilities (translators, managed runs, transcript import) the same way.

Behaviour

None. The same CLI commands and output strings, and the same doctor warnings. CommandRunner is now passed as &mut dyn so the trait can be used as an object.

Testing

• cargo clippy --all-features --all-targets -D warnings is clean.
• cargo test --all-features --no-fail-fast: everything passes, including the new registry test, except trace_runtime::tests::doctor_reports_a_daemon_that_{cannot_authenticate_the_route,resolves_other_credentials}. Those fail the same way on main in my local environment.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-06T12:33:33.369568Z	fd8b9b2	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "Codex (@codex) review" or "Codex (@codex) security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.