Repository navigation
test: T56 — Move scripts/session-e2e.sh off the old session verbs - #3630
Conversation
The session e2e now starts boxes with min session start --detach, runs commands with min box exec, removes boxes with min box rm --force, renames with min box rename, lists stored hooks with min box spec and stops the daemon with min host stop <host>. A one-word exec command runs under bash -c, as the old session exec ran it. The steps that ran a task inside a running session are gone, with a comment that links #3216. The own-IP box stays as the egress control of the task runs. The two NET-058 steps stay on min session policy until #3589, with a marker. A new case, attach_ends_on_stop_and_restart, proves that min box stop and min box restart end an attach with exit 254 and a line that names the cause, and never the shell-exit prompt (#3526, #3547).
scripts_and_cli_reference_use_no_removed_session_verb reads the session e2e, the three smaller scripts and docs/reference/cli*.md. It fails on min session activate, destroy, exec, rename, policy, hooks and run, on min task run --path and on a bare min stop (BCLI-042). A line with the removed-verb-ok marker for #3589 can keep min session policy, for NET-058's two steps. The e2e's last three prose spellings move to the new verbs.
|
Important Review skippedBot user detected. Ask an organization administrator to verify that this bot has an active assigned CodeRabbit review seat. See the seat-assignment guide. After the seat is confirmed, request a review again.
|
|
Reviewed: no findings; approval withheld, 9 threads unresolved at APPROVE withheld: 9 unresolved threads — minimal-foundry[bot] scripts/session-e2e.sh, norrietaylor scripts/session-e2e.sh, norrietaylor scripts/session-e2e.sh, norrietaylor scripts/session-e2e.sh:7147, minimal-foundry[bot] scripts/session-e2e.sh:2313, …. Resolving the last thread releases the approval on this head. Resolve the open threads. Resolving the last one releases the approval on this head; no new review is required. Verdict APPROVE, posted as a comment: the App cannot review its own pull request; the check and the label carry it. Clobber: none seen. |
There was a problem hiding this comment.
arch-review: PASS_WITH_CONDITIONS · 1 condition · the full ruling
The script moves onto the verbs that the design's command tree lists, and the new test guards the removed spellings. The NET-058 marker and the dropped session run step follow spec 28, and each links its issue. The new attach case checks exit 254 and one notice line, as architecture.md line 492 says. The e2e did not run at this head, so that leg is open. C1 asks for a green scripts/session-e2e.sh run on the KVM and macOS lanes, with the links in the PR body.
Outside the diff
scripts/session-e2e.sh:2283: C1 · This case provesarchitecture.mdline 492: exit 254 after one notice line. Nobody ran it at this head. Run it on the KVM and macOS lanes, and put the run links in the PR body. If it fails, fix the step and keep the exit-254 and no-prompt checks.
Ruled at 82512c2 against gominimal/arch@8b824707cde6 by claude-opus-5-5.
The attach can read the box between the restart's stop and its start, and then its line says stopped. The step now retries up to three times and fails when no try prints the restart line, so a restart that always says stopped no longer passes.
|
Round 1 of 10 on the runtime's review findings, pushed as
Verify Converted to a draft: something above needs a person before review. Review threads the coder judged:
No script or CLI reference page uses a removed session spelling, and the test scripts_and_cli_reference_use_no_removed_session_verb fails on any of them. scripts/session-e2e.sh drives the box verbs, and its restart-while-attached step now fails unless the attach prints the line that names the restart. Notes from the agent: This round changes only the restart step in scripts/session-e2e.sh. The step retries up to three times, because the code prints 'was stopped' when its read lands between the restart's stop and start (attach_ending in crates/minimal/src/cmd/boxes/attach.rs). Before each try, the step removes the /home/ae-restart-live marker, so a kept /home does not make the next attach look ready too early. I did not run |
There was a problem hiding this comment.
arch-review: PASS_WITH_CONDITIONS · 1 condition · the full ruling
The new restart loop is stricter than before. Each try still checks exit 254 and no prompt (architecture.md line 492), so the gate holds. C1 is still open. The PR body says that just e2e was not run, and it gives no lane run. Run the e2e on the KVM lane and the macOS lane at this head, and put the two run links under ## Proof.
Outside the diff
scripts/session-e2e.sh:2301: C1 · This loop keeps the exit-254 check of architecture.md line 492, and it fails when no try shows the restart line. No VM run proves it yet. Runattach_ends_on_stop_and_restartand the other cases on the KVM lane and the macOS lane at this head. Put the run links in the PR body.
Ruled at adcff14 against gominimal/arch@8b824707cde6 by claude-opus-5-5.
|
/revise Head The restart step restarts a box that cannot be restarted. The step starts the project's default session entry (
The |
|
That Nothing to type again: the command gets a rocket when its round starts, or a refusal that names the reason. No round was counted for it yet. |
min box restart refuses a box whose spec does not set lifetime = "until_stopped". The attach-end proof started its box from the shared seed's implicit default entry, which sets none, so the native e2e failed at the restart. The proof now seeds its own project with a default session entry that sets the lifetime.
|
Round 2 of 10 on a failed check suite, pushed as
Verify Converted to a draft: something above needs a person before review. No script or CLI reference page uses a removed session spelling, and the test scripts_and_cli_reference_use_no_removed_session_verb enforces this. The attach-end step of scripts/session-e2e.sh now starts its box from an entry that sets lifetime = "until_stopped", so min box restart can restart it and the step can run on the box verbs. Notes from the agent: Revise round 2. The native-daemon-e2e failure was real: the attach-end proof started e2e-attach-end from the shared seed's implicit default entry, which sets no lifetime, so min box restart refused it with exit 2. The proof now seeds its own project (AE_SEED_DIR, removed at the end of the proof and on teardown) with [sessions.default.box] lifetime = "until_stopped". The implicit default entry has an empty body, so the only difference from before is the lifetime. I did not run just e2e on this box, so the e2e lane must confirm the fix. The 'tests' job excerpt in the directive shows only PASS lines and no failed test, so I found no cause to fix there. It can be a follow-on failure or an unrelated one. The ci-linux-native-success aggregator fails because its jobs fail. bash -n and shellcheck show no new findings in the changed region. |
There was a problem hiding this comment.
arch-review: PASS_WITH_CONDITIONS · 1 condition · the full ruling
The new commit fixes the restart step's setup. It seeds a default session entry with lifetime = "until_stopped", and the design says restart follows that attribute (architecture.md line 338). The assertions did not change. The e2e leg is still not run on this head, because the PR body says just e2e was NOT run. C1 stays open. Run the e2e on the KVM and macOS lanes at this head, and link both runs under ## Proof.
Outside the diff
scripts/session-e2e.sh:2295: C1 · This seed is correct.min box restartneedslifetime = "until_stopped"(architecture.md line 338). C1 is still open: no KVM or macOS lane run at this head shows that theattach_ends_on_stop_and_restartcase passes. Run both lanes and link the runs under## Proof.
Ruled at ebb8290 against gominimal/arch@8b824707cde6 by claude-opus-5-5.
The e2e seed now declares [sessions.attach-end] with lifetime = "until_stopped", and the attach-end proof starts that entry in the shared project, so min box restart can restart its box. This replaces the separate seed from the previous round. The seed also declares an empty [sessions.default]. Without it, a start that names no entry picks the only declared session entry, and every other box in the script would get the until_stopped lifetime.
|
/revise Head Drop the restart half of the attach-end proof.
PR body. Under Residuals, add: the restart-while-attached step is in T62 (#3635), which fixes the restart line on VM hosts. |
|
That Nothing to type again: the command gets a rocket when its round starts, or a refusal that names the reason. No round was counted for it yet. |
|
Round 3 of 10 on a /revise note, pushed as
Verify Converted to a draft: something above needs a person before review. No script or CLI reference page uses a removed spelling (the test passes). The attach-end proof in scripts/session-e2e.sh now starts a box from the seed's [sessions.attach-end] entry, which sets lifetime = "until_stopped", so min box restart can restart it. The proof uses the same box for the stop half and the restart half. It asserts the same two Detached lines and keeps the retry on a 'was stopped' line. Notes from the agent: Revise round 3. The branch head (ebb8290) already had a round-2 fix: a separate seed with [sessions.default.box] lifetime = "until_stopped". This round replaces it with the note's design. The shared seed now declares [sessions.attach-end] with lifetime = "until_stopped", and the proof runs |
| proof_attach_ends_on_stop() { | ||
| # The proof starts a box from the project that this script seeds. A | ||
| # project that E2E_PROJECT_DIR names and that brings its own minimal.toml | ||
| # can declare entries that the start cannot pick, so skip it there. | ||
| if [ -z "$SEED_DIR" ] && [ -z "$SEEDED_MFILE" ]; then | ||
| echo "::notice::attach-end proof skipped: $PROJECT_DIR has its own minimal.toml, and this proof runs only in a project that the script seeds" | ||
| return 0 | ||
| fi | ||
| echo "::group::attach-end proof (min box stop ends an attach)" | ||
| ae_name="e2e-attach-end" | ||
| # shellcheck disable=SC2086 | ||
| out="$(cd "$PROJECT_DIR" && mnl session start --detach --name "$ae_name" ${E2E_ACTIVATE_ARGS:-} 2>"$WORK/ae-start.err")" || { | ||
| echo "::error::'min session start --detach --name $ae_name' failed for the attach-end proof" | ||
| echo "--- stderr ---"; cat "$WORK/ae-start.err" 2>/dev/null || true | ||
| fail | ||
| } | ||
| ae_sid="$(printf '%s\n' "$out" | tail -n1 | tr -d '\r')" | ||
|
|
||
| # T62 (gominimal/minimal#3635) adds the restart-while-attached step. | ||
|
|
||
| # T57: a stop ends the attach with exit 254 and one line that names the stop. | ||
| attach_end_start stop "$ae_name" | ||
| mnl box stop "$ae_name" </dev/null >"$WORK/ae-stop-cmd.out" 2>&1 || { | ||
| echo "::error::'min box stop $ae_name' failed while a terminal was attached" | ||
| cat "$WORK/ae-stop-cmd.out" 2>/dev/null || true | ||
| kill "$AE_PID" 2>/dev/null || true | ||
| fail | ||
| } | ||
| attach_end_wait stop "$ae_name" | ||
| ae_lines="$(grep -cF "Detached: $ae_name was stopped." "$WORK/ae-stop.out" || true)" | ||
| if [ "$ae_lines" != 1 ]; then | ||
| echo "::error::the attach that the stop ended printed 'Detached: $ae_name was stopped.' $ae_lines times (expected once)" | ||
| echo "--- transcript ---"; cat "$WORK/ae-stop.out" 2>/dev/null || true | ||
| fail | ||
| fi | ||
| echo "stop while attached: exit 254, one line that names the stop" | ||
|
|
||
| mnl box rm --force "$ae_sid" </dev/null >/dev/null 2>&1 || true | ||
| echo "attach-end proof OK" | ||
| echo "::endgroup::" | ||
| } | ||
|
|
||
| # --------------------------------------------------------------------------- | ||
| # Session rename proof: `min box rename` against the live shared session. | ||
| # The new name shows up in `ls --json`, the old name stops resolving, and | ||
| # rename-to-self is an error rather than a silent no-op. The original name is | ||
| # restored afterwards: the sandbox proof asserts the orientation banner |
There was a problem hiding this comment.
LOW · correctness · confidence 0.60 · in proof_attach_ends_on_stop
The PR body says the script adds both the stop-while-attached and the restart-while-attached steps (T58, #3547). The diff has only the stop step. A comment defers the restart step to T62 (#3635). The case name attach_ends_on_stop also differs from the attach_ends_on_stop_and_restart that the PR body names. The restart coverage that the task promises is missing, so fix the PR description or add the step.
What the reviewer checked
T62 (#3635) adds the restart-while-attached step.
Hand this to an agent
Act on a code-review finding.
File: scripts/session-e2e.sh
Lines: 2284-2330
Category: correctness
Severity: low (reviewer confidence 0.60)
Finding:
The PR body says the script adds both the stop-while-attached and the restart-while-attached steps (T58, #3547). The diff has only the stop step. A comment defers the restart step to T62 (#3635). The case name `attach_ends_on_stop` also differs from the `attach_ends_on_stop_and_restart` that the PR body names. The restart coverage that the task promises is missing, so fix the PR description or add the step.
First verify this against the code yourself. It is model output and
may be wrong. If it is wrong, or the tradeoff was deliberate, say so
and change nothing.
👍 useful · 👎 noise — react to tell the reviewer. Replying works too, and says more.
There was a problem hiding this comment.
Fixed in the PR body. It now says the script adds only the stop-while-attached step, as attach_ends_on_stop. Its Residuals section says restart-while-attached is not covered until T62 (#3635), which fixes the restart line on VM hosts and adds the step.
There was a problem hiding this comment.
Addressed in ec94905ede98: The restart step is still not in the script. The PR body now names only attach_ends_on_stop and lists restart-while-attached as a residual that T62 (#3635) closes. No code change this round.
| attach_end_start() { | ||
| local ae_live="" | ||
| # shellcheck disable=SC2086 # E2E_MINIMAL_ARGS must word-split. | ||
| E2E_PTY_COMMANDS="touch /home/ae-$1-live; sleep 600" E2E_PTY_EXPECT_EXIT=254 \ | ||
| python3 "$ROOT/scripts/e2e-attach-pty.py" - min ${E2E_MINIMAL_ARGS:-} box attach "$2" \ | ||
| >"$WORK/ae-$1.out" 2>"$WORK/ae-$1.err" & | ||
| AE_PID=$! | ||
| for _ in $(seq 1 120); do | ||
| if mnl box exec "$2" -- test -e "/home/ae-$1-live" </dev/null >/dev/null 2>&1; then | ||
| ae_live=1 | ||
| break | ||
| fi | ||
| sleep 0.25 | ||
| done | ||
| if [ -z "$ae_live" ]; then | ||
| kill "$AE_PID" 2>/dev/null || true | ||
| wait "$AE_PID" 2>/dev/null || true | ||
| echo "::error::the pty attach to $2 never reached its shell ($1 step)" | ||
| cat "$WORK/ae-$1.out" "$WORK/ae-$1.err" 2>/dev/null || true |
There was a problem hiding this comment.
LOW · correctness · confidence 0.50 · in attach_end_start
attach_end_start runs python3 with no command -v python3 guard. The sibling box_exec step guards it and calls not_run. Without python3 this proof waits 30 s and fails with 'never reached its shell', which hides the real cause. Also, the failure paths after session start (stop failure, wait failure, wrong line count) call fail without min box rm --force "$ae_sid". The e2e-attach-end box stays behind and a later run can hit the name.
What the reviewer checked
attach_end_start: python3 "$ROOT/scripts/e2e-attach-pty.py" - min ... box attach "$2" (no guard); box rm --force only on the success path
Hand this to an agent
Act on a code-review finding.
File: scripts/session-e2e.sh
Lines: 2244-2262
Category: correctness
Severity: low (reviewer confidence 0.50)
Finding:
`attach_end_start` runs python3 with no `command -v python3` guard. The sibling `box_exec` step guards it and calls `not_run`. Without python3 this proof waits 30 s and fails with 'never reached its shell', which hides the real cause. Also, the failure paths after `session start` (stop failure, wait failure, wrong line count) call `fail` without `min box rm --force "$ae_sid"`. The `e2e-attach-end` box stays behind and a later run can hit the name.
First verify this against the code yourself. It is model output and
may be wrong. If it is wrong, or the tradeoff was deliberate, say so
and change nothing.
👍 useful · 👎 noise — react to tell the reviewer. Replying works too, and says more.
There was a problem hiding this comment.
Deferred as a low residual: gominimal/inbox#1290, listed under Residuals. T62 (#3635) edits this proof next and picks it up there.
There was a problem hiding this comment.
Addressed in ec94905ede98: The proof now reports NOT RUN when python3 is missing, and each failure after the start calls ae_fail, which removes e2e-attach-end before it fails.
|
Round 7 of 10 on a /revise note, pushed as
Verify Outside the task's
Converted to a draft: something above needs a person before review. Review threads the coder judged:
1 thread was already answered in earlier rounds, with nothing new on it since. No script, CLI reference page or justfile line that a person runs uses a removed spelling. The guard now reads the justfile too. The justfile's own-IP demo prints min session start in place of min session activate. This round did not run just e2e. Notes from the agent: Items 1 and 2 of the note, and the --path separator part of item 3, were already on the branch in aad244d (rename to attach_ends_on_stop, wrapper scope closes at }, ${MIN} matched, unit cases for each, --path read per command). This round adds item 3's rest in 563231d. justfile is outside the task's layers. The /revise note asked for that change, so it is in scope for this round. justfile:893 now prints |
There was a problem hiding this comment.
arch-review: PASS_WITH_CONDITIONS · 2 conditions · the full ruling
The new commit adds the justfile to the removed-verb guard, and moves the just up echo to min session start. That conforms to architecture.md line 173. C1 is still open: the PR body says that just e2e was not run, and no lane run links exist for this head. C2 is still open: the body still claims the restart-while-attached step for #3547, but the script proves only the stop. Edit the PR body as C2 says, and add the KVM and macOS run links for this head as C1 says.
Outside the diff
scripts/session-e2e.sh:2284: C1 · This case proves the stop for #3526. Run it on the KVM lane and on the macOS lane at this head. It must pass without theattach-end proof skippednotice. Then put the two run links under## Proofin the PR body.
Ruled at 563231d against gominimal/arch@8b824707cde6 by claude-opus-5-5.
| } | ||
| ae_sid="$(printf '%s\n' "$out" | tail -n1 | tr -d '\r')" | ||
|
|
||
| # T62 (gominimal/minimal#3635) adds the restart-while-attached step. |
There was a problem hiding this comment.
LOW · correctness · confidence 0.60 · in proof_attach_ends_on_stop
The PR says the script adds a restart-while-attached step for T58 (#3547). The new attach_ends_on_stop case only tests stop. The restart step is deferred to T62 (#3635) in a comment. The case name and the PR body claim more than the case checks, so a regression in restart-while-attached passes just e2e. State the gap in the PR, or add the step.
What the reviewer checked
session-e2e.sh:2302 # T62 (#3635) adds the restart-while-attached step.
Hand this to an agent
Act on a code-review finding.
File: scripts/session-e2e.sh
Lines: 2302-2302
Category: correctness
Severity: low (reviewer confidence 0.60)
Finding:
The PR says the script adds a restart-while-attached step for T58 (#3547). The new `attach_ends_on_stop` case only tests stop. The restart step is deferred to T62 (#3635) in a comment. The case name and the PR body claim more than the case checks, so a regression in restart-while-attached passes `just e2e`. State the gap in the PR, or add the step.
First verify this against the code yourself. It is model output and
may be wrong. If it is wrong, or the tradeoff was deliberate, say so
and change nothing.
👍 useful · 👎 noise — react to tell the reviewer. Replying works too, and says more.
There was a problem hiding this comment.
Fixed in the PR body. It now says the script adds only the stop-while-attached step, as attach_ends_on_stop. Its Residuals section says restart-while-attached is not covered until T62 (#3635), which fixes the restart line on VM hosts and adds the step.
There was a problem hiding this comment.
Addressed in ec94905ede98: Same gap as the other restart thread: the PR body states it, and T62 (#3635) adds the restart step. No code change this round.
|
/arch-review |
There was a problem hiding this comment.
arch-review: PASS_WITH_CONDITIONS · 1 condition · the full ruling
The change moves the session e2e and the justfile onto the verbs that the design's command tree lists (architecture.md line 410). It replaces bare min stop with min host stop (line 555). The new guard test refuses each removed spelling. Five NET-058 lines keep min session policy behind a marker until #3589 lands. C1 removes those lines and the marker, and can stay open at merge. The dropped session run and changed-files steps are residuals that #3216 and #3638 own.
Outside the diff
scripts/session-e2e.sh:32178: C1 · This line and the four other marked lines still callmin session policy. Move them to the box verb that #3589 adds, and remove theremoved-verb-okmarker.
Ruled at 563231d against gominimal/arch@8b824707cde6 by claude-opus-5-5.
| /// The marker that lets one line of `scripts/session-e2e.sh` keep | ||
| /// `min session policy`. NET-058's two steps keep the verb until a box verb | ||
| /// gives NET-058's refusal for a box on a stopped VM (gominimal/minimal#3589). | ||
| const REMOVED_VERB_MARKER: &str = "removed-verb-ok: gominimal/minimal#3589"; |
There was a problem hiding this comment.
C1 · This marker lets five lines keep min session policy, which the command tree (architecture.md line 410) does not list. When #3589 lands, delete this constant and its continue branch. Then the test refuses the verb on every line.
There was a problem hiding this comment.
Deferred as a low residual, as the ruling allows: gominimal/inbox#1291. When #3628 (T60, #3589) merges, the five lines move to the box verb, and the removed-verb-ok constant and its continue branch go.
There was a problem hiding this comment.
Addressed in ec94905ede98: #3589 landed on main and the merge moved both NET-058 steps to min box show --network. REMOVED_VERB_MARKER, its continue branch and its test case are deleted, so the guard refuses min session policy on every line.
…ts-session-e2e-sh-off-the-old-session # Conflicts: # scripts/session-e2e.sh
Main moved the NET-058 steps to min box show --network (#3589), so no script line needs the removed-verb-ok marker any more. The guard now refuses min session policy on every line.
The attach-end proof now reports NOT RUN when python3 is missing, in place of a 30 s wait and a misleading failure. Each failure after the box starts removes e2e-attach-end, so a later run can reuse the name.
|
A merge of main before round 8, not counted against the cap (1 of 3), pushed as Resolved conflicts in:
Verify Outside the task's
Converted to a draft: something above needs a person before review. Review threads the coder judged:
No script or CLI reference page uses a removed spelling. The guard test now refuses Notes from the agent: Conflict round: both hunks in scripts/session-e2e.sh took main's side. Main moved the NET-058 unknown-box and stopped-VM steps to |
|
Undrafted: round 8's agent-verify failure is |
|
/arch-review |
There was a problem hiding this comment.
arch-review: PASS · the full ruling
C1 is met. The NET-058 steps in scripts/session-e2e.sh now use min box show --network. The guard test refuses min session policy on every line, and the marker is gone. The command tree in architecture.md (lines 410 and 414) keeps only the box verbs and the session aliases that the script now uses. The restart-while-attached step stays a filed residual on #3635.
Ruled at ec94905 against gominimal/arch@8b824707cde6 by claude-opus-5-5.
There was a problem hiding this comment.
minimal-review — approval withheld
Verdict COMMENT — APPROVE withheld: 9 unresolved threads — minimal-foundry[bot] scripts/session-e2e.sh, norrietaylor scripts/session-e2e.sh, norrietaylor scripts/session-e2e.sh, norrietaylor scripts/session-e2e.sh:7147, minimal-foundry[bot] scripts/session-e2e.sh:2313, …. No findings.
Panel: 3 of 3 personas ran (correctness, security, supply-chain).
|
Re-queued at the same approved head |
Conflict in scripts/session-e2e.sh: main (#3630) replaced the session-run step that followed this branch's task box-stop proof. Kept the box-stop proof and took main's replacement step after it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016CYvTvep9ZaSf5pJMB2hQA
T56 — Move scripts/session-e2e.sh off the old session verbs
slice: S8
kind: task
covers: BCLI-042
size: L
est: 1000 code, 40 tests
executor: agent
difficulty: routine
model: standard
diagnostics: none new
observability: none new
layers:
- migration:
- scripts/session-e2e.sh: moves off
min session activate|destroy|exec|rename|hooks|run,min task run --pathand baremin stop, onto the box verbs and the type nouns; drops each in-sessionsession runstep with a comment linking #3216; adds the stop-while-attached step for T57 (#3526), as the caseattach_ends_on_stop- crates/minimal/src/tests.rs: adds
scripts_and_cli_reference_use_no_removed_session_verb, which reads the four scripts and docs/reference/cli*.md and fails on any removed spelling (min session activate|destroy|exec|rename|policy|hooks|run,min task run --path, baremin stop)test: cargo nextest run -p minimal scripts_and_cli_reference_use_no_removed_session_verb
produces 8000 : none
consumes: none
proof: No script or CLI reference page uses a removed spelling, and
just e2epasses on the box verbs.depends on:
- blocked by #3217
- blocked by #3524
- blocked by #3526
- blocked by #3547
Proof
A person can run the new test. It passes on this branch, so the four scripts and every docs/reference/cli*.md page use none of the removed spellings. The one exception is the NET-058 policy steps, which carry an explicit marker. scripts/session-e2e.sh now drives min session start --detach, min box exec, min box rm --force, min box rename, min box spec and min host stop. It has a new attach_ends_on_stop_and_restart case for #3526 and #3547.
just e2ewas NOT run: this box has no /dev/kvm. So the second half of the proof (just e2epasses on the box verbs) is still owed by the KVM or macOS lane.just e2eat head563231d2, passed on all three lanes:test-kvm): https://github.com/gominimal/minimal/actions/runs/44483516629/job/135425421577e2e): https://github.com/gominimal/minimal/actions/runs/44483516491/job/135429141183. Its log printsattach-end proof (min box stop ends an attach)thenattach-end proof OK.native-daemon-e2e): https://github.com/gominimal/minimal/actions/runs/44483516592/job/135421474115Notes from the agent (from the first round, before CI ran
just e2e): 1.just e2ewas not run, because this box has no /dev/kvm. I ran single cases of the native lane (env -u E2E_VM ./scripts/session-e2e.sh <case>). lifecycle and session_rename passed. session_exec, box_exec, attach_ends_on_stop_and_restart, task_run and hooks failed on this box because no box can start a process here: 'Minimal cannot read the box's process namespace'. The unchanged script fails session_exec the same way here. restart failed only at its on_destroy hook-output check, and the unchanged script fails that same check here. In the new script, min host stop, the min box spec listing of on_destroy and min box rm --force all passed before that check. Treat the e2e evidence as owed by the KVM and macOS lanes.2. Conflict in the task: the test must fail on
min session policy, but the spec keeps NET-058's two policy steps on that verb until #3589 (T60). Those five lines carry the comment# removed-verb-ok: gominimal/minimal#3589. The test allowssession policyonly on a line with that marker. T31 or T60 must remove the marker when it moves the steps.3. Mapping choices:
session activate .becamesession start --detach(the start runs from the project dir).--no-promptbecame the global--no-input, which the activation path reads the same way. A one-wordsession exec <box> '<cmd>'became `box exec <box…[truncated]Tests
cargo nextest run -p minimal scripts_and_cli_reference_use_no_removed_session_verbVerify
min task run agent-verifyexited 100, so this PR is a draft.Residuals
just e2edoes not cover restart-while-attached.attach_end_starthas no python3 guard, and its failure paths leave the box running: gominimal/inbox#1290.Closes #3525