You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@maka.apache.org by "ggbdpq (via GitHub)" <gi...@apache.org> on 2026/09/01 20:32:48 UTC
[PR] fix(cli): exit 1 when a denied sandbox boundary is never retried [maka]
ggbdpq opened a new pull request, #4506:
URL: https://github.com/apache/maka/pull/4506
## Summary
- Non-interactive `maka run` exited `0` after an auto-denied sandbox boundary request whenever any later tool succeeded in a different Step: `TurnOutcomeClassifier` cleared the sole unresolved sandbox failure based on Step IDs alone, so a denied boundary followed by an unrelated tool success settled the Turn as `recovered` (#4388).
- The wire has no retry identity, so recovery stays a heuristic, but a later success can only prove recovery when it retries the same blocked tool in a later step. The classifier now records the failed call's tool name alongside its Step and requires the later success to match it.
- Unrelated successes now leave the failure unresolved: exit `1`, boundary diagnostic on stderr, denied request still available for a later `--continue`.
- Regression coverage: live-event and durable-message classifier cases plus a process-level exit-code case. All three were verified to fail on unpatched `main` before the fix (red) and pass after (green).
## Verification
| Claim | Command | Result |
| --- | --- | --- |
| Target suite green, incl. 3 new `#4388` cases | `node --test dist/__tests__/runtime-host-run-command.test.js` (packages/cli) | 37 tests, 37 pass, 0 fail (2 of the new cases fail on unpatched main) |
| Repo format | `npm run format:check` | Checked 1854 files, no issues |
| ASF headers | `npm run check:asf-headers` | Every source file carries the ASF header or a reviewed exclusion |
| Full cli suite on this Windows machine | `node --test "dist/**/*.test.js"` | 725 tests: 681 pass, 41 fail, 3 skipped |
| The 41 failures are pre-existing platform gaps, not this change | Stashed the change, rebuilt, re-ran two representative failing files, compared with the change applied | Identical both ways: 9 tests / 5 pass / 4 fail. Failures are macOS LaunchAgent plist path assertions, systemd deployment suites, and SIGINT/SIGTERM exit-code tests; none of those files import the changed module |
## AI use
Analysis, patch, and tests were produced with GLM-5.3-Flash (ZCode) under the contributor's direction; the contributor reviewed and is the human contributor of record.
## Checklist
- [x] Tests and checks pass locally (pre-existing Windows-only failures documented and evidenced above)
- [ ] Behavior change: **Yes** - non-interactive runs whose denied boundary is never retried now exit `1` instead of `0`
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For additional commands, e-mail: commits-help@maka.apache.org
Re: [PR] fix(cli): exit 1 when a denied sandbox boundary is never retried [maka]
Posted by "ggbdpq (via GitHub)" <gi...@apache.org>.
ggbdpq commented on PR #4506:
URL: https://github.com/apache/maka/pull/4506#issuecomment-5540849914
The red `test` job on `0796f1112` is not the code: all 798 unit tests pass (795 pass + 3 skips, 0 fail), and the step that failed is the CLI production dependency audit — `npm audit --omit=dev --workspace maka-agent --json` — which threw with an **empty** error payload (`{"summary":"","detail":""}`). That empty payload is npm's registry-audit-endpoint failure signature; a real advisory would carry a populated summary.
I ran the exact same command against the same tree locally: exit `0`, `metadata.vulnerabilities.total = 0`, no error field. The commit also touches no dependency manifests (only `run-command-core.ts` and two test files), so a new advisory appearing between the green run on `28f9b88e` and this run is unlikely.
Could someone re-run the failed job? If the audit still fails on a re-run with a populated advisory report, that is a different conversation — but an empty `{summary, detail}` points at the npm audit endpoint, not at this diff.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For additional commands, e-mail: commits-help@maka.apache.org
Re: [PR] fix(cli): exit 1 when a denied sandbox boundary is never retried [maka]
Posted by "Astro-Han (via GitHub)" <gi...@apache.org>.
Astro-Han commented on code in PR #4506:
URL: https://github.com/apache/maka/pull/4506#discussion_r3921300675
##########
packages/cli/src/runtime-host-run-command.ts:
##########
@@ -584,17 +585,45 @@ type TurnOutcomeObservation =
type TerminalOutcomeObservation = Extract<TurnOutcomeObservation, { kind: 'terminal' }>;
+/**
+ * Canonical serialization for comparing a retry's semantic target against the
+ * blocked call's. Key order is normalized so live and durable observations of
+ * the same call produce identical text. Args that cannot be serialized (or are
+ * missing) yield a unique unprovable marker that never matches anything,
+ * keeping the failure unresolved (fail-closed).
+ */
+let unprovableArgsSeq = 0;
Review Comment:
This counter is now redundant with the guard you added on line 688: return `undefined` for unprovable args and `failedArgsJson !== undefined` already keeps the failure unresolved, while a provable success can never equal `undefined`. That also drops a module-level mutable that makes `stableArgsJson` non-idempotent. And `@maka/core/tool-args-identity` already exports `stableJsonStringify`, which is what the rest of the repo uses to say "same call" and which throws on the values you are catching here, so the whole helper reduces to a try/catch around it.
##########
packages/cli/src/__tests__/runtime-host-run-command.test.ts:
##########
@@ -313,6 +313,82 @@ describe('Runtime Host maka run adapter', () => {
);
});
+ test('returns exit code 1 when a same-named tool retries a different target (live)', async () => {
+ const fixture = runFixture({
+ turnEvents: sandboxBoundaryEvents(
+ 'turn-1',
+ 'step-1',
+ 'step-2',
+ 'Partial answer',
+ 'Read',
+ { path: '/workspace/README.md' },
+ { path: '/outside/secret.txt' },
+ ),
+ });
+
+ const exitCode = await runFixtureCommand(fixture, ['accept same-name different target']);
+
+ assert.equal(exitCode, 1);
+ });
+
+ test('returns exit code 1 when a same-named tool retries a different target (durable)', async () => {
+ const fixture = runFixture({
+ graph: true,
+ finalMessages: sandboxBoundaryMessages(
+ 'step-1',
+ 'step-2',
+ 'Read',
+ { path: '/workspace/README.md' },
+ { path: '/outside/secret.txt' },
+ ),
+ });
+
+ const exitCode = await runFixtureCommand(fixture, [
+ 'accept durable same-name different target',
+ '--graph',
+ ]);
+
+ assert.equal(exitCode, 1);
+ });
+
+ test('still reports recovered when the retry targets the same semantic args (live)', async () => {
Review Comment:
This asserts a live recovery the wire cannot produce: `session-projector.ts:616` sets `args: undefined` on every projected `tool_start`, so both calls serialize to distinct `unprovable:N` and the real run exits 1. It passes because `toolStart` defaults to `args: {}`. If `recovered` stays, this case needs the projector's shape (`args: undefined` plus `argsPreview`).
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For additional commands, e-mail: commits-help@maka.apache.org
Re: [PR] fix(cli): exit 1 when a denied sandbox boundary is never retried [maka]
Posted by "Astro-Han (via GitHub)" <gi...@apache.org>.
Astro-Han commented on code in PR #4506:
URL: https://github.com/apache/maka/pull/4506#discussion_r3927071090
##########
packages/cli/src/__tests__/activation-command.test.ts:
##########
@@ -461,6 +461,57 @@ describe('maka activate JSONL protocol', () => {
});
});
+ test('blocks a completed invocation whose stream carried a boundary failure', async () => {
+ // Before #4506 the classifier cleared `sandboxBoundary` when a later tool
+ // succeeded, so this activation completed with exit 0. The stream event is
Review Comment:
P2: this comment, and "A new regression pins this transition" in the body, are not true of this fixture. `completedResult()` sets `sandboxBoundary: 'none'` (`:80-87`), so on `main` the old condition evaluates `streamBoundaryFailure && ('none' !== 'recovered')`, which is already true, and the activation already returned `blocked` / `permission_required` / 3. The new condition `streamBoundaryFailure || ...` gives the same answer. The case passes on both sides and pins nothing.
The only shape that actually changes needs `invocation.sandboxBoundary === 'recovered'`, and that value no longer exists in the type, so a fixture that hands `MakaRunOutcome` straight to `fakeDeps` cannot express it. To be clear, this is the shape I asked for last round, and my suggestion had the same hole, so this is not on you to have caught.
Smallest fix: reword the comment and the body to say the case guards the current contract, and state plainly that the `maka activate` transition is not regression-coverable after the deletion. If you would rather cover it for real, the fixture has to drive the actual classifier instead of injecting the outcome.
##########
packages/cli/src/run-command-core.ts:
##########
@@ -311,10 +311,7 @@ export async function runMakaTextCliCore(
...(parsed.options.hostProfileId ? { hostProfileId: parsed.options.hostProfileId } : {}),
...(parsed.options.projectId ? { projectId: parsed.options.projectId } : {}),
runOutcomeObserver: (result) => {
- if (result.sandboxBoundary === 'recovered') {
- boundaryFailureInvocationIds.delete(result.outcomeId);
- unclassifiedBoundaryFailure = false;
- } else if (result.sandboxBoundary === 'unresolved') {
+ if (result.sandboxBoundary === 'unresolved') {
boundaryFailureInvocationIds.add(result.outcomeId);
unclassifiedBoundaryFailure = false;
Review Comment:
P3: this assignment is dead after the `recovered` branch is gone. Whenever it runs, line 315 has just added an id, and the only reader is `unclassifiedBoundaryFailure || boundaryFailureInvocationIds.size > 0` at `:446`, so the result is the same either way. It existed to pair with the branch you deleted.
While you are here: nothing reads the ids in `boundaryFailureInvocationIds`, only `.size > 0`, so the two variables collapse into one boolean. Roughly five more lines, no assertion changes, and this change is what made them redundant.
##########
packages/cli/src/__tests__/runtime-host-run-command.test.ts:
##########
@@ -1673,14 +1742,15 @@ function storedToolCall(
stepId: string,
ts: number,
toolName = 'Read',
+ args: unknown = {},
Review Comment:
P3: this parameter has no caller. Every `storedToolCall` site in the file leaves it at the default. The matching parameter on `toolStart` is passed only by `deniedWideningEvents` (`:1573`, `:1577`), where the two path strings are decoration because the classifier no longer reads args. `:1413` and `:1559` also pass `'Read'` explicitly, which is already the default. Four no-op additions inside a PR whose point is deletion; suggest reverting all of them.
##########
packages/cli/src/__tests__/runtime-host-run-command.test.ts:
##########
@@ -313,6 +313,50 @@ describe('Runtime Host maka run adapter', () => {
);
});
+ test('returns exit code 1 when a same-named tool retries a different target (live)', async () => {
Review Comment:
P3: this name and the one at `:326` say "retries a different target", but the fixture has no targets. `sandboxBoundaryEvents` and `sandboxBoundaryMessages` never pass args, so both the failing and the succeeding call get the helper default `{}` (`:1559`, `:1566`, `:1413`, `:1418`), and the classifier reads no args at all now. The names claim a discrimination the code does not have, which is the same thing I flagged last round about the fixtures. Renaming to "keeps the boundary unresolved when a later same-named call succeeds (live)" / "(durable)" fixes it.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For additional commands, e-mail: commits-help@maka.apache.org
Re: [PR] fix(cli): exit 1 when a denied sandbox boundary is never retried [maka]
Posted by "yihanzhu (via GitHub)" <gi...@apache.org>.
yihanzhu commented on PR #4506:
URL: https://github.com/apache/maka/pull/4506#issuecomment-5517660217
On the Host path the denial never reaches the exit decision: `#observeTurn` drops `sandbox_boundary_request` before run-command-core sees it, so after a denied widening a `Read` of a workspace file passes the name-only check now on main and the run exits 0 with the apology on stdout. Since a denied widening can't be retried non-interactively, any `request_sandbox_boundary` call in the Turn could simply pin it to `unresolved`, no args needed. The trade-off is that it also fails a Turn that over-declared, got denied, then narrowed the declaration and succeeded, which seems acceptable for an exit code; the alternative is to feed the actual denial from the interaction bridge into the classifier, but that doesn't cover the durable path. I have red/green tests for the false-success shape in the projector's real `argsPreview` form if useful.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For additional commands, e-mail: commits-help@maka.apache.org
Re: [PR] fix(cli): exit 1 when a denied sandbox boundary is never retried [maka]
Posted by "ggbdpq (via GitHub)" <gi...@apache.org>.
ggbdpq commented on PR #4506:
URL: https://github.com/apache/maka/pull/4506#issuecomment-5508959542
Fixed in 6ee47cd3d — the recovery correlation now compares the retried call's semantic target, not just its name:
- Both live `tool_start` events and durable `tool_call` messages carry `args`, so the classifier serializes each call's args with sorted keys (`stableArgsJson`) and requires an exact match with the blocked call's args before treating a later success as recovery of that boundary.
- Missing or unserializable args produce a unique unprovable marker that never matches anything, so the failure stays unresolved (fail-closed), per the review's "if the wire data cannot prove it, keep it unresolved".
- Regressions added: live and durable fixtures with two same-named `Read` calls whose `args.path` differ (`/outside/secret.txt` blocked, `/workspace/README.md` later succeeds) — both now exit 1 — plus a same-target retry fixture that must still report `recovered` and exit 0.
Local: target suite 40/40 pass on Windows; the full CLI suite shows only the pre-existing platform failures (LaunchAgent/EPERM fsync/symlink) also present without this change.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For additional commands, e-mail: commits-help@maka.apache.org
Re: [PR] fix(cli): exit 1 when a denied sandbox boundary is never retried [maka]
Posted by "Astro-Han (via GitHub)" <gi...@apache.org>.
Astro-Han commented on code in PR #4506:
URL: https://github.com/apache/maka/pull/4506#discussion_r3915441028
##########
packages/cli/src/runtime-host-run-command.ts:
##########
@@ -584,15 +585,46 @@ type TurnOutcomeObservation =
type TerminalOutcomeObservation = Extract<TurnOutcomeObservation, { kind: 'terminal' }>;
+/**
+ * Canonical serialization for comparing a retry's semantic target against the
+ * blocked call's. Key order is normalized so live and durable observations of
+ * the same call produce identical text. Args that cannot be serialized (or are
+ * missing) yield a unique unprovable marker that never matches anything,
+ * keeping the failure unresolved (fail-closed).
+ */
+let unprovableArgsSeq = 0;
+
Review Comment:
On the live path this branch is the only one ever taken: `session-projector.ts:619` sets `args: undefined` on every `tool_start` (`session-continuity.ts:69` says live frames carry only `argsPreview`). So `recovered` is unreachable in production and a genuinely widened retry still exits 1. Also `@maka/core/tool-args-identity` already has `stableJsonStringify`, and `undefined` works as the sentinel because of the guard on line 770.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For additional commands, e-mail: commits-help@maka.apache.org
Re: [PR] fix(cli): exit 1 when a denied sandbox boundary is never retried [maka]
Posted by "ggbdpq (via GitHub)" <gi...@apache.org>.
ggbdpq commented on PR #4506:
URL: https://github.com/apache/maka/pull/4506#issuecomment-5518985510
Pushed 8d8d482dc addressing both follow-ups:
- Your Host-path gap: `request_sandbox_boundary` tool_starts now pin the failure set for the rest of the Turn, so once a widening is requested no later success (including the projector's `argsPreview`-shaped false-success you reproduced) can downgrade the exit to 0. The interaction event itself stays dropped, but the widening call's `tool_start` carries the pin, so the durable replay path is covered by the same rule. Your trade-off note is taken as-is: over-declare-then-narrow now also fails the exit code.
- Rebase note: main absorbed the original name-only fix while this PR waited, so the branch was rebased to carry only the incremental args-correlation commit (jackwener's P1) plus this pin.
Red/green would have been welcome - the added regressions cover the three shapes: live and durable same-name/different-target (exit 1), same-target retry still recovers (exit 0), and denied-widening-then-success (exit 1 with the diagnostic). CI is running on 8d8d482dc.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For additional commands, e-mail: commits-help@maka.apache.org
Re: [PR] fix(cli): exit 1 when a denied sandbox boundary is never retried [maka]
Posted by "likun666661 (via GitHub)" <gi...@apache.org>.
likun666661 merged PR #4506:
URL: https://github.com/apache/maka/pull/4506
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For additional commands, e-mail: commits-help@maka.apache.org
Re: [PR] fix(cli): exit 1 when a denied sandbox boundary is never retried [maka]
Posted by "ggbdpq (via GitHub)" <gi...@apache.org>.
ggbdpq commented on PR #4506:
URL: https://github.com/apache/maka/pull/4506#issuecomment-5526713539
All four review items are addressed, pushed in d617b59bb on top of the net deletion:
- **P1 (body)**: the body is rewritten. It now states the change is a deliberate tightening with explicit compatibility impact — `maka run` unresolved exits are the pinned common path (exit `1`, empty stdout, diagnostic on stderr; the flipped root-Graph fixture pins the `0 → 1` transition), and `maka activate` returns `blocked` / `permission_required` / exit `3` where it previously completed with exit `0`. The old "exit codes are unchanged" claim and the args/target reasoning that belonged to the dropped version are gone; the surviving argument is the one that supports the deletion: the condition compared information that cannot establish recovery, and non-interactive runs deny widening unconditionally, so real recovery does not exist on this surface.
- **P2 (write-only mechanism)**: `#unresolvedSandboxFailures` is now a `Set<string>` of blocked tool-use ids; `#callByToolUseId`, the `tool_call` observation variant, and the step/name projections in both observation producers are deleted.
- **P2 (activation regression)**: `blocks a completed invocation whose stream carried a boundary failure` pins the completed → `blocked` / `permission_required` / exit `3` transition.
- **P2 (empty stdout)**: stated in the body as the visible consequence, with the write-the-explanation follow-up named; happy to fold it in here if you prefer.
- **P3 (test collapse)**: the args threading is deleted from both fixtures and the same-target live case is removed — the different-target live case, the durable case, and `deniedWideningEvents` (with its stderr assertion) cover the tightened behavior.
- **P3 (red-on-base count)**: the body now names the SHA the earlier red count was measured against (`main@6e6af952a`, pre-#4389) and states that per-case counts against the moved base (`9d4002b38`, which includes #4389) were not re-measured.
Suites: `runtime-host-run-command` 38/38, `activation-command` 12/12 (new regression included). The manual acceptance from the review (a real `maka run` whose second call reads an in-workspace file) still needs a live Host, so it remains unexecuted — the fixture pair covers the same decision points statically.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For additional commands, e-mail: commits-help@maka.apache.org
Re: [PR] fix(cli): exit 1 when a denied sandbox boundary is never retried [maka]
Posted by "ggbdpq (via GitHub)" <gi...@apache.org>.
ggbdpq commented on PR #4506:
URL: https://github.com/apache/maka/pull/4506#issuecomment-5537902395
All four items addressed in 0796f1112:
**P2 (body)**: the activation section now states the fixture guards the current contract rather than pinning the transition, and calls out plainly that the transition itself is not regression-coverable after the deletion (since `recovered` no longer exists in the outcome type).
**P2 (activation test comment)**: reworded to match — the comment explains the fixture guards the contract and that an injected `MakaRunOutcome` cannot express the old shape.
**P3 (run-command-core)**: `boundaryFailureInvocationIds` and `unclassifiedBoundaryFailure` collapsed into a single `boundaryFailure` boolean. The set's only reader was `.size > 0`, and the dead assignment at `:316` paired with the deleted `recovered` branch. Five lines removed, no assertion changes.
**P3 (test names and params)**: the two "retries a different target" cases renamed to "keeps the boundary unresolved when a later same-named call succeeds (live/durable)"; reverted the `args` parameter additions on `toolStart` and `storedToolCall` (four call sites). The classifier reads no args after the deletion, so the path strings were decoration.
Verification: runtime-host-run-command 39/39, activation-command 12/12, format:check clean, asf-headers clean.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For additional commands, e-mail: commits-help@maka.apache.org
Re: [PR] fix(cli): exit 1 when a denied sandbox boundary is never retried [maka]
Posted by "ggbdpq (via GitHub)" <gi...@apache.org>.
ggbdpq commented on PR #4506:
URL: https://github.com/apache/maka/pull/4506#issuecomment-5521519979
Took your reading - the branch is now a net deletion, pushed as 2c49da979 (single commit after the rebase; the args-correlation and pin commits were dropped since the deletion subsumes them):
- `sandboxBoundary` collapses to `none | unresolved`. `stableArgsJson`, the `argsJson` plumbing, the pin, the `recovered` clause in `run-command-core`, and the `!== 'recovered'` guard in `activation-command.ts:592` are all gone. Exit codes are unchanged on every path you reached.
- Both named follow-ups acknowledged in the PR body: the dead `sandbox_boundary_request` branch in `run-command-core` (with its fixture scenario rework) and the unresolved-exit suppression of the Turn's final message. I did not touch either here so the PR stays a pure deletion.
- Tests: the live "same-target retry recovers" case you flagged now asserts the projector's real shape can only exit 1; the live/durable same-name/different-target cases and the denied-widening case stay as permanent-unresolved regressions; "recovered from live and durable Turns" is flipped to assert `unresolved` on both.
Target suite 39/39 on Windows; Biome clean on the four changed files. CI is running on 2c49da979.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For additional commands, e-mail: commits-help@maka.apache.org