Repository navigation
PASS_WITH_CONDITIONS
Inputs. First ruling on #3630 at 82512c2c, on arch 8b824707. No since.diff, since-conditions.json, findings.json or arch-map.md is staged, so I found the sections by search. The task covers BCLI-042 (docs/specs/28-spec-box-cli, line 536, with the T31 rollout bullets at lines 672 to 683).
Governs. architecture.md Command tree (lines 410 to 414): the session noun has only list | start | attach | logs | stop | rm, and the box-wide verbs live under min box. Type nouns (line 458): session start --detach is the TTY-less create, and exec, rename and spec live under min box. min box attach (line 492) and Exit codes (line 376): an attach that a stop ends returns 254 after a one-line notice on stderr.
Ruling. The move conforms. Each removed spelling in scripts/session-e2e.sh maps to a verb that the command tree lists. The two other e2e scripts, kernel-bump-review.sh and every docs/reference/cli*.md page are already clean at this head (I searched them). The test scripts_and_cli_reference_use_no_removed_session_verb sits beside cli_reference_has_no_retired_commands in crates/minimal/src/tests.rs, which is the right owner. The removed-verb-ok: gominimal/minimal#3589 marker is an honest interim: it names its issue and its flip condition (T60), and the test admits only policy on a marked line. The dropped session run step links #3216, which spec 28 names as the residual. In the restart step, a was stopped line is a warning, not a failure. A restart begins with a stop, and line 492 lists a stop as a cause for 254, so this does not relax the gate. Exit 254 and the no-prompt check stay hard. The one open item is rubric rule 4: just e2e did not run at this head, and the task's proof needs it. C1 closes it on the KVM and macOS lanes, which always run on a PR.
Precedent. No ledger entry decides a caller migration off removed verbs, so none is cited. #3628 (T60 work) is related, but its why does not state a decision this ruling reuses.
For T31, not this change. architecture.md line 458 still says min session activate --sync "stays on its alias (gominimal/arch#161, owner ruling 2026-11-06)". Spec 28 (Non-goals, line 605) records a later owner ruling (2026-11-09) that no alias keeps it. This script never passes --sync, and no reference page names it, so the conflict does not decide this change. T31 removes the verb, and its ruling must settle this conflict with an arch amendment.
Not checked. I did not build or run the tests. I did not confirm that local0 and local-minvmd0 are the names min host list prints on each lane; the e2e run in C1 settles that. The PR's agent-verify failure is minimal::otel_cli exec_run_attach_and_task_run_share_the_commands_trace, which this diff does not touch. It is a code-quality matter for minimal-review.
Conditions
- C1: Run
scripts/session-e2e.shat head82512c2c812e01cbbac9fe2f7f69fde0ef02158con the KVM lane (ci-linux-kvm) and the macOS lane, with the caseslifecycle,session_exec,box_exec,attach_ends_on_stop_and_restart,session_rename,task_runandhooksgreen. Then replace the PR body's line "just e2ewas NOT run" under## Proofwith the two lane run links. Until then the e2e leg is not_run, and it is open. If a case fails, fix the script step that it names, not the assertion:attach_end_waitkeeps its exit-254 and no-prompt checks.
Ruled at 82512c2 against gominimal/arch@8b824707cde6 by claude-opus-5-5; arch_rulings 1421, job e0959647-ab7c-4b3b-93d9-a041d7ba51a1.