Exit codes are consistent cross-backend w/ structured ScriptResponse JSON by theelliotm · Pull Request #1336 · microsoft/mxc

Elliot (theelliotm) · 2026-09-29T23:41:11Z

📖 Description

A backend that refused a policy exited -1 with a generic backend_error, so a caller could not
tell a refused policy from a crash, a launch failure or a timeout.

FailurePhase now derives the wire error code, and the executor envelope and the engine share
that one mapping:

phase	code	exit
`Rejected`	`policy_validation`	1
`BackendUnavailable`	`backend_unavailable`	-1
everything else	`backend_error`	-1

Every ScriptResponse failure site across Bubblewrap, LXC, Seatbelt, process container,
IsolationSession, Nanvix, Hyperlight, and Windows Sandbox is tagged with the phase that matches its cause:

rejected — the request itself is refusable; only a changed request can succeed.
unavailable — the host cannot run this backend, so a caller may fall back a tier.
error — a runtime failure (spawn, syscall, IO, crash, invariant).

Genuinely mixed cases stay error. Mislabelling infrastructure as policy_validation is the
failure this contract exists to prevent.

From review:

Rejected is policy-only. Windows Sandbox mapped SandboxUnavailable onto it, reporting a
disabled optional feature as policy_validation; it now uses BackendUnavailable.
emit_captured_stderr no longer suppresses a workload's own stderr. Windows Sandbox mirrors a
failing guest's stderr into error_message, so the de-duplication swallowed real output; it now
applies only to phases where no workload ran.
validate_legacy_host_lists refusals are tagged, matching the rest of the shared validator.
captured_stderr_to_emit is pure, so the tests assert the exact bytes emitted for both the
suppressed and the relayed case.

P13b was a false green: its inheritDefaultEnv=false case supplied a verbatim env block missing
SYSTEMROOT/LOCALAPPDATA, which was always refused — the -1 exit hid it. The fixture now uses
Get-MinimalEnv, so the pair varies only by the field under test.

Exit-code expectations move from -1 to 1 where a refusal is now typed:
wslc_destroy_on_exit_false_rejected, two IsolationSession one-shot cases, and the Node Seatbelt
blockedHosts test (backend_error -> policy_validation).

Concistency Fixes

The mapping above only bound the surfaces the bug touched. A sweep of every backend found six
paths that still disagreed about what a failure means:

Executor entry. A failure raised before a ScriptResponse exists printed a bare stderr line
and exited 1 — indistinguishable from a policy refusal. wxc, lxc and mxc_darwin now emit a
typed envelope through emit_mxc_error_exit.
IsolationSession timeout. One-shot never applied with_service_timeout_grace, so the service
timer won the race and a timeout arrived as an ordinary exit 1, colliding with the rejection code.
create_process now returns ExecOutcome with a split deadline, matching piped_exec_handle.
nanvix / hyperlight. A single Preflight variant collapsed policy refusals, missing shipped
artifacts and environment faults into one code; each splits into Policy / Unavailable / Setup.
Spent deadlines. wslc, LXC and Windows Sandbox reported a watchdog kill as the workload's own
exit. LXC gains a typed AttachError and Windows Sandbox a Termination tag on ExecResult, so
the kill survives the call instead of being flattened into a sentinel.
Availability. Explicitly selecting LXC skipped the probe discovery runs; macOS claimed
Seatbelt without checking sandbox-exec. Both now fail as backend_unavailable up front.
Feature gates. A disabled build feature is backend_unavailable; a missing experimental
opt-in is malformed_request. These were previously swapped in places.

🔗 References

Contributes to #612.

🔍 Validation

cargo fmt --all -- --check and cargo clippy --workspace --all-targets -- -D warnings are clean.
cargo test on the touched crates passes against a stashed baseline: wxc_common 905/49 and
windows_sandbox_lifecycle 184/0 each gain exactly the one test added here, and
process_container_common matches baseline at 315/1. isolation_session (195), lxc_common (355),
mxc_engine (153), nanvix_runner (44) and wslc_common (263) are fully green. The remaining
local failures are pre-existing and environment-dependent (telemetry/registry, PSEC RPC,
reparse-point tests) in modules this PR does not touch. Seatbelt and hyperlight do not build on
this host, so their changes are covered by CI. Host suites run separately (P8a/P8c/P8d/P8e, P13b).

✅ Checklist

Signed the Contributor License Agreement
Linked to an issue
Updated documentation (if applicable)
Updated Copilot instructions (if build, architecture, or conventions changed)
If this PR changes Cargo.lock, the dependency-feed-check check passes (see docs/pull-requests.md)

📋 Issue Type

Bug fix
Feature
Task

Microsoft Reviewers: Open in CodeFlow

Copilot

Copilot review overview

🟡 Changes recommended

Shared rejection and stderr handling currently misclassifies an unavailable backend and suppresses workload stderr.

Review effort: Balanced
Findings: 2

Open (2)

Classify unavailable backends as infrastructure failures, not policy rejection · New
Preserve guest stderr for ordinary ProcessExited responses · New

What changed in this PR

This PR adds typed policy-rejection handling across one-shot executors.

Changes:

Maps rejection phases to policy_validation and exit code 1.
De-duplicates and newline-terminates captured stderr.
Updates validators, fixtures, tests, and backend documentation.

File	Description
`tests/scripts/run_wslc_all_tests.ps1`	Expects exit 1 for WSLC policy refusal.
`tests/scripts/run_processcontainer_lifecycle_test.ps1`	Supplies a valid minimal environment.
`src/core/wxc_common/src/validator.rs`	Classifies shared validation failures as rejections.
`src/core/wxc_common/src/script_runner.rs`	Centralizes exit-code and error-envelope handling.
`src/core/mxc_engine/src/dispatch.rs`	Reuses phase-to-error-code mapping.
`src/core/mxc_darwin/src/main.rs`	Uses shared stderr and exit helpers.
`src/core/lxc/src/main.rs`	Uses shared stderr and exit helpers.
`docs/process-container/networking.md`	Documents typed ProcessContainer refusals.
`docs/isolation-session/oneshot.md`	Clarifies one-shot refusal classifications.

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

…-policy-rejections-bug # Conflicts: # src/backends/process_container/common/src/base_container_runner.rs

Copilot

Copilot review overview

🟡 Changes recommended

Global rejection mapping misclassifies Windows Sandbox unavailability, and some shared validator failures remain untagged.

Review effort: Balanced
Findings: 1 · 3 · 2

Open (6)

Map host-unavailability failures to BackendUnavailable · New
Retag legacy host-list validation errors as policy rejections · New
Preserve guest stderr for ordinary ProcessExited responses
Classify unavailable backends as infrastructure failures, not policy rejection
Document -1 for all backend and infrastructure failures · New
Assert exact stderr output in both test branches · New

Audit every ScriptResponse failure-construction site across the backends and tag each with the phase that matches its cause, so the wire error code derived from FailurePhase is truthful: - rejected: the request itself is refusable; only a changed request succeeds. - unavailable: the host cannot run this backend, so a caller may fall back. - error: a runtime failure (spawn, syscall, IO, crash, invariant). Covers bubblewrap, lxc, seatbelt, process_container, isolation_session, and windows_sandbox. Genuinely mixed cases stay as error, since mislabelling infrastructure as policy_validation is the failure this contract exists to prevent. Also addresses PR review: - FailurePhase::Rejected is now policy-only. Windows Sandbox mapped SandboxUnavailable onto it, which reported a disabled optional feature as policy_validation; it now uses BackendUnavailable. - emit_captured_stderr no longer suppresses a workload's own stderr. Windows Sandbox mirrors a failing guest's stderr into error_message, so the de-duplication swallowed real output; it now applies only to phases where no workload ran. - validate_legacy_host_lists refusals are tagged as rejections, matching the rest of the shared validator. - captured_stderr_to_emit is pure, so the tests assert the exact bytes emitted for both the suppressed and relayed cases. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Copilot review overview

🟡 Changes recommended

Seatbelt currently classifies the same invalid working directory differently by launch method, and stderr relay introduces an avoidable unbounded copy.

Review effort: Balanced
Findings: 1 · 2 · 3

Open (6)

Map host-unavailability failures to BackendUnavailable
Validate working directories consistently across launch methods · New
Avoid duplicating captured stderr in memory · New
Clarify crash exit code versus MXC failure behavior · New
Document propagation of workload exit codes · New
Correct rustdoc for workload exit code preservation · New

Resolved since last review (5)

Retag legacy host-list validation errors as policy rejections
Preserve guest stderr for ordinary ProcessExited responses
Classify unavailable backends as infrastructure failures, not policy rejection
Assert exact stderr output in both test branches
Document -1 for all backend and infrastructure failures

Copilot

Copilot review overview

🔵 Needs a closer look

Dry-run errors lack envelopes, stderr bytes can be altered, and one Seatbelt policy failure remains misclassified.

Review effort: Balanced
Findings: None

Resolved since last review (6)

Map host-unavailability failures to BackendUnavailable
Avoid duplicating captured stderr in memory
Validate working directories consistently across launch methods
Correct rustdoc for workload exit code preservation
Document propagation of workload exit codes
Clarify crash exit code versus MXC failure behavior

Previously missed (3)

In code that hasn't changed since last review

Classify embedded NUL profile errors as policy validation

src/backends/seatbelt/common/src/seatbelt_runner.rs:648

An embedded NUL can only come from the caller-supplied/generated profile (not from a runtime failure), so this still reports a policy refusal as backend_error. A seatbelt.profileOverride containing \u0000, for example, can only succeed after the request changes and should produce policy_validation under the new contract.

Emit policy validation envelope before rejected dry-run exit

src/core/wxc_common/src/script_runner.rs:69

A rejected dry run now exits with 1, but this function terminates before every executor reaches emit_backend_error_envelope. Because validation itself does not write the rejection reason to the logger, dry-run callers receive neither the promised policy_validation envelope nor the reason. Emit the response stderr/envelope here before exiting (the helpers are no-ops for a successful dry run).

Avoid appending newline when relaying workload stderr

src/core/wxc_common/src/script_runner.rs:115

This appends a newline to captured workload stderr even when no diagnostic follows. For example, a successful LXC workload that writes progress without a newline has error_message empty, so the old eprint! preserved progress but this helper emits progress\n. Only request an added newline when envelope_applies(response); otherwise relay the workload bytes unchanged, and update the helper test accordingly.

- A rejected dry run exited 1 with neither the reason nor its envelope: every executor calls handle_dry_run_exit before reaching the relay helpers, and it diverges. It now emits both before exiting; each is a no-op for a dry run that passed. - Captured workload stderr is relayed exactly as written. The terminating newline is added only when the error envelope follows and would otherwise start mid-line, so unterminated progress output no longer gains a newline it never wrote. - An embedded NUL in a Seatbelt profile is a rejection: the profile is built from the request, verbatim for seatbelt.profileOverride, so only a changed request can succeed. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Copilot review overview

🔵 Needs a closer look

Some structural and BaseContainer failures remain misclassified, and post-launch stderr can still be suppressed.

Review effort: Balanced
Findings: None

Previously missed (3)

In code that hasn't changed since last review

Move policy checks before backend availability and dry-run branches

src/backends/process_container/common/src/base_container_runner.rs:1178

This rejection is not reached for BaseContainerRunner::new(): on an API-capable host, can_backend_service_request returns PolicyIncompatible for least_privilege_mode, so the preceding branch reports backend_unavailable; dry-run also returns before this check. Move the caller-fixable policy check ahead of both branches so direct and dry-run callers receive policy_validation.

Preserve workload stderr after post-launch relay failures

src/core/wxc_common/src/script_runner.rs:100

PostLaunchFailed explicitly includes failures “while running user code” (models.rs:1314-1317), so this predicate can say the workload did not run after it already emitted stderr. If a relay failure preserves and mirrors that partial stderr into error_message, captured_stderr_to_emit drops the workload bytes. The response needs an explicit stderr-origin/workload-started signal (and diagnostic constructors should use it) rather than inferring this from an ambiguous phase.

Classify empty commands as malformed, not policy validation

src/core/wxc_common/src/validator.rs:215

Rejected is documented as a policy refusal, but an empty command is structural malformed input. The state-aware equivalent already returns malformed_request in validator.rs:246-250; this path will instead emit policy_validation. Keep this unclassified until ScriptResponse can represent malformed input, rather than mislabeling it as policy.

…cy-rejections-bug

BaseContainerRunner::validate ran the leastPrivilege refusal after both the dry-run return and the serviceability probe, so a caller-fixable policy was reported as backend_unavailable on an API-capable host and not reported at all on a dry run. BaseContainerRequestDecision already separates PolicyIncompatible from the host-capability verdicts; honour that split instead of collapsing every non-serviceable decision into unavailable. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Copilot review overview

🟡 Changes recommended

The IsolationSession environment-refusal host test still expects exit -1 even though the new classification guarantees exit 1.

Review effort: Balanced
Findings: 1

Open (1)

Update host test expectation for reclassified isolation session exit code · New

Both backends funnelled every error into a ScriptResponse carrying only exit -1, so a watchdog kill was indistinguishable from a spawn failure -- the inference FailurePhase exists to prevent. Derive the phase from each error variant at the single to_response funnel. Hyperlight folded RunError::TimedOut into a generic Runtime error, losing the timeout outright; give it a Timeout variant that renders identical text. Preflight stays unclassified in both: it mixes missing host artifacts with refused policy, so it cannot honestly claim either phase. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Copilot review overview

🟡 Changes recommended

Seatbelt policy-build failures remain untyped, and Hyperlight setup failures are incorrectly classified as post-launch failures.

Review effort: Balanced
Findings: 1 · 1

Open (2)

Update host test expectation for reclassified isolation session exit code
Distinguish Hyperlight launch failures from post-launch runtime errors · New

Previously missed (1)

In code that hasn't changed since last review

Map Seatbelt policy rejections to typed rejected responses

src/backends/seatbelt/common/src/seatbelt_runner.rs:180

A caller-supplied Seatbelt filesystem path containing .. is rejected only inside build_profile_with_proxy (profile_builder.rs:729-755), but this maps every builder failure to an unclassified ScriptResponse::error. That request therefore still exits -1 with backend_error, contrary to the new rejection contract. Because this builder also reports host/environment failures such as a missing HOME, preserve those as backend errors while giving policy-build failures a typed result that maps to ScriptResponse::rejected.

reject_unhonorable_environment raises IsolationSessionError::Policy, which now exits 1 like the ui and lifecycle refusals beside it. The expectation was left at -1 when those two were updated, so the case would have failed deterministically on a host run. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Copilot review overview

🟡 Changes recommended

Rejection status reporting remains inconsistent, and audit post-processing can misclassify infrastructure failures as policy validation.

Review effort: Balanced
Findings: 1 · 2

Open (3)

Update host test expectation for reclassified isolation session exit code
Rejections retain -1 exit code in telemetry and result reporting · New
Distinguish Hyperlight launch failures from post-launch runtime errors

Previously missed (1)

In code that hasn't changed since last review

Audit finalization errors retain incorrect Rejected failure phase

src/core/wxc/src/main.rs:1777

With --audit, audit::finalize can fail after the runner returned Rejected; the error branch at lines 1686-1694 changes exit_code and error_message but leaves failure_phase as Rejected. This mapping then exits 1 (and the preceding envelope emits policy_validation) even though the final failure includes audit infrastructure. Reset the phase to an infrastructure phase when audit finalization fails so this mixed case remains backend_error/-1.

lxc_runner.rs conflicted wholesale because main normalized it from CRLF to LF (#1334) while this branch changed its content. Main made no content change to the file, so the resolution keeps this branch's rejection classifications and adopts main's LF line endings.

A rejection reached the OS exit code only through process_exit_code, so display_script_results and completion telemetry read -1 from the response while the process exited 1. FailurePhase::mxc_exit_code is now the single source for the code MXC reports for its own failures, and every producer that builds a Rejected response uses it. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

The phase-to-code mapping landed earlier only bound the surfaces the bug touched. Six paths still disagreed about what a given failure means. - Executors that fail before a ScriptResponse exists now emit a typed envelope via emit_mxc_error_exit instead of a bare stderr line + exit 1. - IsolationSession one-shot never applied the service-timeout grace, so the service timer won the race and a timeout surfaced as exit 1 -- colliding with the rejection code. create_process now returns ExecOutcome with a split deadline, matching piped_exec_handle. - nanvix and hyperlight collapsed policy refusals, missing artifacts and environment faults into one Preflight variant; each now splits into Policy / Unavailable / Setup. - wslc, lxc and windows_sandbox reported a spent deadline as an ordinary workload exit. LXC gains a typed AttachError and windows_sandbox a Termination tag on ExecResult so the kill survives the call. - Explicit LXC selection now runs the same probe discovery uses, and macOS validates sandbox-exec before claiming the backend. - A disabled feature is backend_unavailable; a missing experimental opt-in is malformed_request. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

…-policy-rejections-bug

Copilot

Copilot review overview

🟡 Changes recommended

Several paths still misclassify failures or emit duplicate and unstructured stderr.

Review effort: Balanced
Findings: 6

Open (6)

Classify host filesystem failures as setup errors · New
Keep timeout diagnostics out of standard error · New
Avoid duplicating timeout diagnostics in standard error · New
Structure buffered diagnostics instead of printing raw stderr · New
Classify empty commands as malformed requests · New
Distinguish Hyperlight launch failures from post-launch runtime errors

Resolved since last review (2)

Update host test expectation for reclassified isolation session exit code
Rejections retain -1 exit code in telemetry and result reporting

The in-process sandbox handle merged from main carried two failure sites that predate the classification contract. - Refusing StdioMode::Inherit is a refusal, not a fault: only a changed call can succeed, and it does so on every host. IsolationSession already returns `rejected` for the identical condition. - A non-Linux host cannot launch an LXC sandbox at all, which is a host prerequisite rather than a bad request, so it reports backend_unavailable and a caller may fall back a tier. The two junction deny tests asserted the pre-contract exit code. They were passing only because their `mklink /J` guard was skipping them on an unprivileged host; both now run and assert the phase. Document the constructor-to-code-to-exit mapping in the Copilot instructions so a new failure site is classified rather than defaulted. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

- A hyperlight mount path whose parent is missing is host-filesystem state, not a refusal: the message itself says `mkdir -p` fixes it, so the same request succeeds elsewhere. It moves from Policy to Setup. - The IsolationSession and LXC timeout branches built their diagnostic with ScriptResponse::error, which copies the message into standard_err. Because a timed-out workload did run, the executor relays standard_err bare and the envelope then printed the same text again. Both now keep the diagnostic in error_message only, matching the timeout responses in mxc_engine and sandbox_process. - emit_mxc_error_exit wrote the buffered log to stderr, ahead of the envelope, leaving stderr unparseable whenever anything had been logged. The buffer goes to stdout, as handle_dry_run_exit already does. - An empty command reported policy_validation from the one-shot path and malformed_request from the state-aware one. FailurePhase cannot gain a variant -- it crosses the WSLc daemon IPC boundary, where a cached image may predate the host -- so ScriptResponse carries an optional code that outranks the phase, and the two surfaces now agree. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Copilot review overview

🟡 Changes recommended

Several timeout and failure-classification paths remain incorrect or inconsistent with the stated cross-backend contract.

Review effort: Balanced
Findings: 6

Open (6)

Avoid relying on timer ordering to classify expired deadlines · New
Classify mixed working-directory failures as backend errors · New
Preserve guest timeout reason instead of labeling it ProcessExited · New
Map disabled backends to backend_unavailable consistently · New
Use wire error code for one-shot completion telemetry · New
Track stderr provenance instead of inferring it from failure phase · New

Resolved since last review (6)

Classify empty commands as malformed requests
Structure buffered diagnostics instead of printing raw stderr
Avoid duplicating timeout diagnostics in standard error
Keep timeout diagnostics out of standard error
Classify host filesystem failures as setup errors
Distinguish Hyperlight launch failures from post-launch runtime errors

+/// can opt in. A backend this build was not compiled with, or whose host
+/// prerequisite is missing, returns a
+/// [`backend_unavailable`](MxcError::backend_unavailable) error; one that
+/// cannot run on this platform at all returns an


+    pub fn malformed(msg: &str) -> Self {
+        ScriptResponse {
+            error_code: Some(MxcErrorCode::MalformedRequest),
+            ..Self::rejected(msg)
+        }


…he wire code Every branch of the seatbelt working-directory check reports host state -- a path that is a file, a missing search bit, or an absent directory -- so the same request can succeed once the host is repaired. That is a mixed cause, which the contract keeps as a backend error rather than a policy refusal. Completion telemetry now classifies from a response's explicit wire code when it carries one, so ScriptResponse::malformed records ConfigError to match the state-aware path instead of the phase's generic PolicyError. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Aligning the state-aware gate with run-to-completion changed a missing experimental opt-in from backend_unavailable to malformed_request, but only the engine's own tests were updated. mxc_ffi and mxc-sdk assert the same gate through the C ABI and the Rust SDK, so every platform job failed on six stale expectations. The SDK rustdoc documented the old code as well: the exec_sandbox opt-in refusal, and a TimedOut note that collapsed the opt-in and the feature-off refusals into one code. They are now distinct. Drops two rationales that the alignment made obsolete -- a wslc id no longer shares a code with the gate, and the FFI test can no longer separate "passed the gate" from "parsed the request". Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Copilot review overview

🟡 Changes recommended

Several paths still misclassify host-dependent failures, nonzero workload exits, feature gates, and early-exit telemetry.

Review effort: Balanced
Findings: 6

Open (6)

Keep host-dependent normalization failures as backend errors · New
Retain host-dependent object normalization failures as backend errors · New
Use typed error for telemetry classification and exit code · New
Align telemetry classification and exit code with MxcError · New
Use wire error code for one-shot completion telemetry
Map disabled backends to backend_unavailable consistently

Resolved since last review (4)

Track stderr provenance instead of inferring it from failure phase
Preserve guest timeout reason instead of labeling it ProcessExited
Classify mixed working-directory failures as backend errors
Avoid relying on timer ordering to classify expired deadlines

Previously missed (4)

In code that hasn't changed since last review

Set ProcessExited for nonzero workload exits

src/backends/isolation_session/common/src/one_shot.rs:177

ExecOutcome::Exited also carries nonzero workload exits, but this response defaults every such result to FailurePhase::None. Set ProcessExited when exit_code != 0; otherwise the serialized response says there was no failure even though the workload failed.

Report trusted directory API failures as backend errors

src/backends/process_container/common/src/appcontainer_runner.rs:1536

find_bfscfg_exe() returns Err only when the trusted Windows-directory API itself fails; a missing binary is Ok(None) and is handled separately. Reclassifying this runtime API failure as backend_unavailable incorrectly invites tier fallback instead of reporting the infrastructure fault as backend_error.

Mark nonzero normal exits as ProcessExited

src/backends/wslc/common/src/wsl_container_runner.rs:1330

A normal exit is not always successful: collect_output already has a test passing exit code 42, but this new match labels it FailurePhase::None. That breaks the stated phase contract and prevents consumers from distinguishing a workload failure from success; use ProcessExited for nonzero exited outcomes.

Distinguish disabled x86_64 feature from unsupported architecture

src/core/mxc_engine/src/run.rs:557

This branch combines two different cases and gives an incorrect message on x86_64 builds where only the feature is disabled: the host already is x86_64, yet the error says x86_64 is required. Split feature-disabled x86_64 from unsupported architectures so the former reports how to enable the build feature and the latter remains unsupported_containment as documented above.

            }
            Ok(None) => request,
-            Err(msg) => return Err(ScriptResponse::error(&msg)),
+            Err(msg) => return Err(ScriptResponse::rejected(&msg)),


            }
            Ok(None) => request,
-            Err(msg) => return Err(ScriptResponse::error(&msg)),
+            Err(msg) => return Err(ScriptResponse::rejected(&msg)),


                telemetry::FailureReason::InitError,
            );
-            process::exit(1);
+            emit_mxc_error_exit(&e.to_mxc_error(), &mut logger);


                telemetry::FailureReason::InitError,
            );
-            process::exit(1);
+            emit_mxc_error_exit(&e.to_mxc_error(), &mut logger);


Jeff Whiteside (jsidewhite)

Copilot AI balanced review requested due to automatic review settings September 29, 2026 23:41

Copilot started reviewing on behalf of Elliot (theelliotm) September 29, 2026 23:41 View session

Copilot AI reviewed Sep 29, 2026

View reviewed changes

Comment thread src/core/wxc_common/src/script_runner.rs Outdated

Comment thread src/core/wxc_common/src/script_runner.rs Outdated

Elliot (theelliotm) added 2 commits September 29, 2026 16:45

Fix process container using backend-error for bad policy

fdeb616

Merge remote-tracking branch 'origin/main' into user/emichlin/backend…

fc51815

…-policy-rejections-bug # Conflicts: # src/backends/process_container/common/src/base_container_runner.rs

Elliot (theelliotm) force-pushed the user/emichlin/backend-policy-rejections-bug branch from 6233494 to fc51815 Compare September 29, 2026 23:53

Copilot AI balanced review requested due to automatic review settings September 29, 2026 23:53

Copilot started reviewing on behalf of Elliot (theelliotm) September 29, 2026 23:54 View session

Copilot AI reviewed Sep 29, 2026

View reviewed changes

Comment thread src/core/wxc_common/src/models.rs

Comment thread src/core/wxc_common/src/validator.rs

Comment thread docs/process-container/networking.md Outdated

Comment thread src/core/wxc_common/src/script_runner.rs Outdated

Copilot AI balanced review requested due to automatic review settings September 30, 2026 00:36

Copilot started reviewing on behalf of Elliot (theelliotm) September 30, 2026 00:37 View session

Copilot AI reviewed Sep 30, 2026

View reviewed changes

Addressing copilot feedback

9d3c668

Copilot AI balanced review requested due to automatic review settings September 30, 2026 00:58

Copilot started reviewing on behalf of Elliot (theelliotm) September 30, 2026 00:59 View session

Copilot AI reviewed Sep 30, 2026

View reviewed changes

Copilot AI balanced review requested due to automatic review settings September 30, 2026 01:11

Copilot started reviewing on behalf of Elliot (theelliotm) September 30, 2026 01:11 View session

Copilot AI reviewed Sep 30, 2026

View reviewed changes

Elliot (theelliotm) and others added 2 commits September 30, 2026 10:47

Merge remote-tracking branch 'origin' into user/emichlin/backend-poli…

5f7bd04

…cy-rejections-bug

Copilot AI balanced review requested due to automatic review settings September 30, 2026 17:56

Copilot started reviewing on behalf of Elliot (theelliotm) September 30, 2026 17:57 View session

Copilot AI reviewed Sep 30, 2026

View reviewed changes

Comment thread src/backends/isolation_session/common/src/error.rs

Copilot AI balanced review requested due to automatic review settings September 30, 2026 18:11

Copilot started reviewing on behalf of Elliot (theelliotm) September 30, 2026 18:12 View session

Copilot AI reviewed Sep 30, 2026

View reviewed changes

Comment thread src/backends/hyperlight/common/src/lib.rs Outdated

Copilot AI balanced review requested due to automatic review settings September 30, 2026 18:17

Copilot started reviewing on behalf of Elliot (theelliotm) September 30, 2026 18:18 View session

Copilot AI reviewed Sep 30, 2026

View reviewed changes

Comment thread src/core/wxc_common/src/models.rs

Elliot (theelliotm) and others added 2 commits September 30, 2026 13:16

Elliot (theelliotm) self-assigned this Sep 30, 2026

Copilot AI balanced review requested due to automatic review settings September 30, 2026 21:58

Copilot started reviewing on behalf of Elliot (theelliotm) September 30, 2026 21:59 View session

Merge remote-tracking branch 'origin/main' into user/emichlin/backend…

823418a

…-policy-rejections-bug

Copilot AI reviewed Sep 30, 2026

View reviewed changes

Elliot (theelliotm) changed the title ~~Exit 1 with a typed code when a backend rejects a policy~~ Exit codes are consistent cross-backend w/ structured ScriptResponse JSON Sep 30, 2026

Copilot AI balanced review requested due to automatic review settings September 30, 2026 22:25

microsoft-github-policy-service Bot added the Copilot-Instructions label Sep 30, 2026

Copilot started reviewing on behalf of Elliot (theelliotm) September 30, 2026 22:26 View session

Copilot AI reviewed Sep 30, 2026

View reviewed changes

Elliot (theelliotm) and others added 2 commits September 30, 2026 15:52

Copilot AI balanced review requested due to automatic review settings September 30, 2026 23:15

Copilot started reviewing on behalf of Elliot (theelliotm) September 30, 2026 23:16 View session

Copilot AI reviewed Sep 30, 2026

View reviewed changes

Jeff Whiteside (jsidewhite) approved these changes Oct 1, 2026

View reviewed changes

Conversation

Elliot (theelliotm) commented Sep 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

📖 Description

Concistency Fixes

🔗 References

🔍 Validation

✅ Checklist

📋 Issue Type

Microsoft Reviewers: Open in CodeFlow

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot review overview

🟡 Changes recommended

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot review overview

🟡 Changes recommended

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot review overview

🟡 Changes recommended

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot review overview

🔵 Needs a closer look

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot review overview

🔵 Needs a closer look

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot review overview

🟡 Changes recommended

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot review overview

🟡 Changes recommended

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot review overview

🟡 Changes recommended

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot review overview

🟡 Changes recommended

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot review overview

🟡 Changes recommended

Uh oh!

Elliot (theelliotm) commented Sep 29, 2026 •

edited

Loading