[JacobStephens2]

Summary

Concentrate installed-Harness selection and all six Model/Effort interactions behind one semantic Setup effect.

 Setup effects
- installed-Harness query + six catalog/check operations
+ ask_harness_settings(current) -> settled choice | retention | error

 Harness interaction
+ selection -> discovery/checks -> canonicalization/retries -> complete answer
+ terminal interface: read(prompt), say(line)

Closes #454

Evidence

• Before: registered_order_and_enter_defaults_produce_one_complete_answer failed at the new interaction with Harness selection not yet implemented.
  After: all 27 complete Harness interaction tests pass, including defaults, clearing, distinct validation rules, atomic retention and interruption.

• All 48 Setup effect tests pass. Existing regressions pass: 18 process-tree tests, 17 terminal/command Setup tests and 8 first-offer tests.

• Normal checks: cargo fmt --check and cargo clippy --all-targets -- -D warnings pass. cargo test passes: 1222 tests across 39 test binaries, 0 failures.

• GitHub CI passes on Ubuntu and macOS. Both platform artifacts also pass the release checks.

Test seams

• harness::settings::ask: scripted terminal transcripts and semantic results, with isolated executable/catalog stand-ins and controlled PATH.
• Setup's ask_harness_settings effect: received defaults, complete accepted-result persistence, complete retention and errors before writes.
• Existing command/PTY Setup, first Run offer and owned-process cleanup regressions.

Merge Danger

Door: two-way

An ownership refactor that can be reverted without migrating User config or Credentials.

Blast Radius: Setup

Interactive Setup and the first Run's Setup offer use the deepened Harness interaction.

Unaddressed findings

Standards

1. Call-order/count assertions in the Setup effect test (src/setup.rs): general testing guidance discourages observing collaborator sequences. Kept because Deepen Harness settings behind one Setup interaction #454 explicitly requires a once-only semantic interaction, recorded defaults and effect-order coverage at this agreed seam (Decisions 6 and 15).
2. Possible duplicated default-answer/EOF helpers (src/harness/settings.rs, src/setup/questions.rs): kept these small helpers local to their respective question flows to preserve ownership and the narrow Harness interface; a shared question utility offers limited benefit in this refactor.

Spec

None. The independent Spec review reported no findings.

Built with codex · gpt-6.1-sol · xhigh