[JacobStephens2]

Summary

Clarify whether a Run notification has evidence of the Model used. The email marks configured settings and reported Models, and explains that a launched Harness does not confirm an unreported Model.

-Built with codex · gpt-6.1-sol · high
+Built with codex · gpt-6.1-sol · high (configured)

 Session models:
+Listed sessions launched the harness. Requested models are not confirmed by harness evidence.
-- implement: codex · gpt-6.1-sol (requested) · session effort: high
+- implement: codex · gpt-6.1-sol (requested; actual model not reported) · session effort: high

Closes #662

Evidence

• Before: cargo test --test run_notification a_codex_notification_distinguishes_launched_sessions_from_unconfirmed_models -- --exact failed because the email did not distinguish configuration from confirmation. The reported-Model regression also failed before the (reported) annotation was added.
  After: Both regressions pass, including a review reporting a different Model from the configured one.
• cargo test --test run_notification: 30 passed; cargo test session::recording --bin thirdshift: 12 passed; cargo test --test pickup_run: 60 passed; cargo test --test harness_choice: 10 passed.
• cargo check, cargo fmt --check, and cargo clippy --all-targets -- -D warnings pass.
• The full cargo test run passed every target before site_deploy, where the unchanged systemd check rejected a warning from an unrelated /etc/systemd/system service. The same test passes with the host's custom unit directory excluded. All site_deploy tests and remaining suite targets pass with SYSTEMD_UNIT_PATH=/usr/lib/systemd/system:/lib/systemd/system cargo test --test site_deploy --test spec_run --test toolchain --test update --test user_config. No deployment files changed.

Test seams

• CLI Run notification delivered to the Resend stand-in: launched Codex sessions with unconfirmed requests; reported Claude Models; Harness default without Model evidence.
• Recording's notification_lines interface: attribution survives failures and retained-evidence replacement.

Review

thirdshift-code-review against main: Standards 0 findings; Spec 0 findings. Each axis read all seven changed files. Changed files left unread: none.

Unaddressed findings

Standards

None.

Spec

None.

Merge Danger

Door: two-way

Notification wording and regression expectations; no persisted schema or execution changes.

Blast Radius: notifications

Run notification consumers see the clearer Model annotations and configuration label.

Built with codex · gpt-6.1-sol · high