[JacobStephens2]

Summary

thirdshift architect now takes base-fix / no-base-fix (and their -- forms) and hands them, or the User config's base.fix, to the Spec run or Run its plan is dispatched as. The Architecture review watches no CI, so a Base fix can only start there.

 thirdshift architect [<focus>] [flags]
   Architecture review  ->  plan marked ready-for-agent
   dispatch the plan, as thirdshift <plan URL> would
     goal      = merge | no-merge      else merge.always
     parallel  = parallel <n>          else spec.parallel
+    base fix  = base-fix | no-base-fix else base.fix
+  last lines and the one Run notification carry "Base fix: <issue URL> merged"

 thirdshift architect --plan-only
-  rejects merge, no-merge, parallel
+  rejects merge, no-merge, parallel, base-fix, no-base-fix   (nothing is dispatched)

A Run and an Architect run's dispatched run now end through one place, so "as thirdshift <plan URL> would" isn't kept true by hand twice:

 main / architect
-  BaseFix::new -> run::run(.., &mut base_fix) -> base_fix.report()   # in both
-  run_outcome(&ended, base_fix) ; notification.send(&ended, base_fix)
+  run::run_to_end(.., child, BaseFixAsk) -> run::Ended { outcome, base_fix }
+  run_outcome(&ended) ; notification.send(&ended)

This PR also merges main into the Spec branch's line. The ticket is blocked by #221 (the Base fix), which merged to main but was not on issue-27. The first commit, 85c7939, merges origin/main in; that is most of the diff (the 17 commits of #226). The ticket's own work is the two commits after it:

85c7939  Merge origin/main into issue-234          # #221's Base fix; 5 files conflicted
f12e473  Pass Base fix through an Architect run    # the ticket
f2ac7a1  Address review                            # run::Ended, run_to_end, Spec run test

Merge resolutions worth a look:

• tests/fakes/gh.rs: both sides added gh issue create. Kept main's, which refuses a label the repository lacks, as gh does, and added this branch's createdAt. The Architect run tests now give their repository needs-triage, ready-for-agent and architecture through a scenario() helper.
• src/notification.rs: this branch's Ending gained main's base_fix, so an Architect run's notification has the Base fix: line a Run's has.
• src/args.rs: kept this branch's ask_parallel helper for both parsers, with main's base-fix arms and child Run kinds.

Seams tested: the argument parser (unit tests in src/args.rs), and the thirdshift binary through tests/support::Scenario with the fake claude and gh (tests/architect_run.rs, tests/command_surface.rs).

Evidence

• Before: architect base-fix, with the plan's Run meeting an Inherited failure (base_fix_lets_the_dispatched_run_start_a_base_fix, written first):

  thirdshift: dispatching the plan .../issues/8, as thirdshift .../issues/8 would
  thirdshift: CI failed on 0bc7f8e: test
  thirdshift: Inherited failures (also failing on main at ba3813f): test
  thirdshift: CI red on test, which also fails on main at ba3813f; fix main first
  test result: FAILED. 0 passed; 1 failed
  

  After: the dispatched Run starts Base fix CI watching, CI-fix Repair and the Repair cap #9 into main, merges the fixed main in and ends ready for review.

  cargo test --test architect_run base_fix
  test the_one_notification_says_what_became_of_the_dispatched_runs_base_fix ... ok
  test without_base_fix_the_dispatched_run_fails_on_an_inherited_failure ... ok
  test the_user_configs_base_fix_applies_to_the_dispatched_run_unless_no_base_fix_is_given ... ok
  test base_fix_lets_the_dispatched_run_start_a_base_fix ... ok
  test base_fix_or_the_user_configs_reaches_each_tickets_run_of_the_dispatched_spec_run ... ok
  test result: ok. 5 passed; 0 failed
  

• Before: architect --plan-only base-fix took base-fix as the focus.
  After: exit 2: merge, no-merge, parallel, base-fix and no-base-fix can't be used with --plan-only: it dispatches no run for them to apply to. architect base-fix --no-base-fix: base-fix and no-base-fix can't be used together.
• Full suite after the last commit: cargo test, 572 passed, 0 failed, across 28 test binaries. cargo clippy --all-targets and cargo fmt --check are clean.

Acceptance criteria:

• base-fix and no-base-fix are accepted by architect and rejected together.
• The flag, or the User config's base.fix, reaches the dispatched Spec run or Run.
• base-fix or no-base-fix with --plan-only is rejected.
• The README mentions it, as does thirdshift help.

Unaddressed findings

Reviewed with /thirdshift:code-review against issue-27. Addressed: the dispatch sequence duplicated between main and architect, the outcome-and-Base-fix pair travelling as a tuple, the Spec-publishing script re-inlined in a test, two unwrapped doc comments, and the User config default not being tested on a dispatched Spec run.

Standards

• The base-fix / no-base-fix arms are repeated in parse and parse_architect. Skipped: the file already does this for merge and email, with ask_once as the shared part; the two arms differ only in where they store the ask, and a helper for them alone would break that pattern.
• RED, GREEN, checks_on_head, checks_on_origin and inherited_failure copy helpers in tests/spec_run.rs. Skipped: each test file keeping its own small script helpers is the repository's habit (checks_on_head was already in three files), and moving them to tests/support is a change across test files this ticket doesn't own.
• without_base_fix_the_dispatched_run_fails_on_an_inherited_failure repeats the assertions of the no-base-fix loop. Skipped: they pin different behaviors, the default with nothing set and the flag overriding base.fix = true; either could break without the other.

Spec

• A Base fix for the dispatched Spec run's own Spec PR, into the Base branch, isn't tested through architect. Skipped: it takes the same BaseFixAsk through the same run::run_to_end as the Ticket case that is tested, and the Spec PR's Base fix itself is covered in tests/spec_run.rs.
• The Base fix: line in the Architect run's last lines and notification goes beyond the ticket's literal criteria. Kept, and the reviewer didn't call it scope creep: Architect run: thirdshift finds and implements a Strong deepening opportunity unattended #27 has base-fix applied "as if given to thirdshift <plan URL>" and the dispatched run's notification suppressed, so without it the report would be lost.

Merge Danger

Door: two-way

Reverting the last two commits removes the flags. The merge of main is the part that is awkward to undo, but issue-27 has to take main in before #241 can merge anyway.

Blast Radius: Architect-runs

architect with neither flag and no base.fix behaves as before. With base.fix = true in the User config, an Architect run's dispatched run may now start a Base fix where it couldn't before, which is what the setting says. A word base-fix given to architect is no longer read as the focus.

Closes #234

🤖 Generated with Claude Code