Fix safety-critical bugs and raise test coverage to 76% by wrhalpin · Pull Request #13 · wrhalpin/RedGNAT

wrhalpin · 2026-07-04T23:38:50Z

Summary

A full code analysis surfaced exploitable holes in the safety controls, an unbounded feedback loop, and several scope escapes. This PR fixes all Critical→Medium findings (fail-closed policy) and raises test coverage from 42% → 76%, focusing the new tests on the safety-critical decision points.

All 316 unit tests pass.

Bug fixes

Fail-closed safety controls (Critical)

KillSwitch.is_active() now treats the durable Postgres record as authoritative for the "clear" decision — a Redis cache-miss (eviction/restart) can no longer hide a standing kill — and halts when the durable store is unreachable.
The runner checks safety before every step (including the first and single-step probe runs, which previously bypassed it) and halts on any check error instead of continuing.
Added RunStatus.EXPIRED + run-status aggregation so token-expiry halts are distinct from kills and an all-error run is FAILED, not COMPLETED.

Phase 2 engagement authorization (High/Medium)

POST /engage/authorize now requires the privileged X-Kill-Key (was mintable with only the general API key while revoke required the elevated key), caps duration at 24h, and maps gate errors to 403/422 instead of a 500.
Gate 2 can verify REDGNAT_PHASE2_UNLOCK against a configured phase2_unlock_secret (a real shared secret) instead of presence-only.

Technique scope & safety (High/Medium)

Scope.allows_cidr now requires containment (subnet_of), closing the /8-passes-a-/16 scan escape.
ad_enum reads its LDAP server/credentials from trusted config only (never ctx.params) and validates the host against scope before binding — closes a credential-exfil path. cloud_enum/token_theft gained Entra/Okta tenant scope gates.
Fixed ZeroDivisionError at max_rate_per_minute = 0, the always-dead oauth_abuse domain check, worker-starving blocking sleeps (now capped), missing GoPhish teardown on error, and an unset send_by_date.

STIX validity, intake dedup, probe runaway (High/Medium)

Valid STIX sighting IDs (UUIDv5) and a deterministic gap_id (fixes dangling Grouping→Note refs and duplicate notes).
Deterministic feed/scenario IDs stop unbounded re-ingestion on every poll.
Added a probe-depth runaway guard (feedback.max_probe_depth), bounding the self-amplifying gap→probe→emulate loop.
Investigation evidence push now retries-on-409 (reopen) and logs real failures; the plugin can pull grouping.

Completeness & hygiene

Tactic parent-fallback in the builder, token_theft dead-code + impossible-travel time window, CORS for DELETE/kill, dead-code removal, and lint reduced 92 → 37.

Testing

Coverage 42% → 76% (make coverage gate of 70% now passes), 158 → 316 tests. New tests target the untested layers by mocking each module's external boundary (DB, HTTP, Celery, GNAT/GoPhish clients): the full API layer, ScenarioStore, the runners (including fail-closed paths), Celery tasks, subscribers, config, and the credential-attack safety controls (Control #4, confirm gates, rate=0, scope gates, push-with-409-retry). Several new tests double as regression guards for the fixes above.

Intentionally left uncovered: HTTP/LDAP enumeration mechanics, PDF/DOCX renderers, and the thin GNAT connector wrapper — mechanical, not risk-bearing.

🤖 Generated with Claude Code

Generated by Claude Code

Copilot

Pull request overview

This PR hardens RedGNAT’s safety controls (fail-closed behavior, stricter scope enforcement, bounded feedback/probe loop), improves STIX/GNAT interoperability with deterministic IDs and investigation scoping, and adds substantial unit test coverage across the API, runners, intake, and techniques.

Changes:

Implement fail-closed safety checks (kill switch + engagement gate) and add run/result status handling for killed/expired/error-only runs.
Tighten scope controls (CIDR containment, LDAP host validation, IdP tenant/host gating) and add operational caps (worker sleep bounds, inline GoPhish polling cap, probe-depth runaway guard).
Introduce deterministic UUIDv5 IDs for deduplication/STIX reference stability and expand tests broadly across the stack.

Reviewed changes

Copilot reviewed 74 out of 74 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tests/unit/test_stix_stamping.py	Remove unused imports in STIX stamping tests.
tests/unit/test_push_routing.py	Remove unused datetime import in push routing tests.
tests/unit/test_push_error_handling.py	Remove unused pytest import in push error handling tests.
tests/unit/test_hypothesis_validation.py	Remove unused imports in hypothesis validation tests.
tests/unit/test_grouping_envelope.py	Remove unused pytest import in grouping envelope tests.
tests/unit/test_engagement_investigation_fields.py	Fix imports for investigation-field tests.
tests/unit/test_config.py	New: config property coverage tests.
tests/unit/test_client.py	New: RedGNATClient facade tests with mocked store/tasks.
tests/unit/test_batch_e_fixes.py	New: regression tests for mapper fallback + impossible-travel window.
tests/unit/test_batch_d_fixes.py	New: regression tests for deterministic IDs + probe depth threading.
tests/unit/techniques/test_scope.py	Add CIDR containment regression tests.
tests/unit/techniques/test_scope_gates.py	New: scope-gate tests for AD/cloud/token-theft techniques.
tests/unit/techniques/test_registry.py	Clean up imports in registry tests.
tests/unit/techniques/test_phishing_success.py	New: GoPhish technique success + teardown-on-error tests.
tests/unit/techniques/test_credential_safety.py	New: Control #4 safety tests for credential-access techniques.
tests/unit/techniques/test_batch_c_fixes.py	New: tests for rate-delay safety + OAuth domain extraction.
tests/unit/scenarios/test_store.py	New: ScenarioStore tests with mocked psycopg connection.
tests/unit/orm/test_models.py	Strengthen STIX sighting ID validity assertions.
tests/unit/intake/test_subscribers.py	New: subscriber polling/health tests with mocked network boundary.
tests/unit/feedback/test_push_internals.py	New: investigation push internals + 409 reopen retry tests.
tests/unit/feedback/test_probe_generator.py	Remove unused pytest import in probe generator tests.
tests/unit/feedback/test_gap_reporter.py	Remove unused pytest import in gap reporter tests.
tests/unit/engagement/test_token.py	Update UTC usage + import cleanup in token tests.
tests/unit/engagement/test_kill_switch.py	Add fail-closed Redis/Postgres behavior tests.
tests/unit/engagement/test_gate.py	Add Gate 2 shared-secret matching tests; UTC updates.
tests/unit/emulation/test_tasks.py	New: Celery task-body tests (feedback/probes/engagement gate).
tests/unit/emulation/test_runner.py	New: runner fail-closed + status aggregation tests.
tests/unit/emulation/init.py	New: package marker/license header.
tests/unit/api/test_routes.py	New: FastAPI route tests for scenarios/runs/intel/STIX.
tests/unit/api/test_engage_routes.py	New: engagement route tests (authorize/revoke/kill/reset).
tests/unit/api/conftest.py	New: shared API test fixtures + fake client/store wiring.
tests/unit/api/init.py	New: package marker/license header.
tests/conftest.py	Remove unused imports from shared test fixtures.
redgnat/techniques/registry.py	Reorder/clarify registry imports; improve typing + alias docs.
redgnat/techniques/phishing/spearphishing_link.py	Add send_by_date, bounded inline polling, teardown on error.
redgnat/techniques/phishing/spearphishing_attachment.py	Bound inline polling; teardown on error.
redgnat/techniques/phishing/mfa_phishing.py	Bound inline polling; teardown on error.
redgnat/techniques/phishing/base.py	Add best-effort GoPhish teardown helper.
redgnat/techniques/identity/token_theft.py	Add scope gates for IdP hosts + improve impossible-travel detection.
redgnat/techniques/identity/password_spray.py	Use safe rate-delay helper to avoid divide-by-zero.
redgnat/techniques/identity/oauth_abuse.py	Fix domain extraction; bound polling; teardown on error.
redgnat/techniques/identity/credential_stuffing.py	Use safe rate-delay helper to avoid divide-by-zero.
redgnat/techniques/identity/base.py	Add `_rate_delay_seconds` helper to harden rate limiting.
redgnat/techniques/discovery/network_scan.py	Remove unused imports.
redgnat/techniques/discovery/cloud_enum.py	Add scope gates for Entra/Okta tenant/host + improve ImportError chaining.
redgnat/techniques/discovery/ad_enum.py	Enforce trusted-config LDAP server/creds + host-in-scope validation.
redgnat/techniques/base.py	Tighten CIDR allowance (containment) + UTC handling updates.
redgnat/scenarios/ttp_mapper.py	Add parent fallback for subtechnique name/tactic.
redgnat/scenarios/builder.py	Use fallback-aware mapper lookups when building plans.
redgnat/reports/cart_report.py	Switch to datetime.UTC usage.
redgnat/plugins/gnat_plugin.py	Support grouping objects + simplify endpoint routing.
redgnat/orm/models.py	Add RunStatus.EXPIRED; make STIX sighting refs deterministic/valid.
redgnat/orm/base.py	Add UUIDv5 deterministic_id + UTC now helper update.
redgnat/intake/sandgnat_subscriber.py	Make SandGNAT feed_id deterministic (dedupe).
redgnat/intake/normalizer.py	Make scenario_id deterministic from feed_id (dedupe).
redgnat/intake/gnat_subscriber.py	Make GNAT feed_id deterministic (dedupe).
redgnat/feedback/probe_generator.py	Add probe depth field + UTC handling updates.
redgnat/feedback/investigation_context.py	UTC handling update for grouping timestamps.
redgnat/feedback/gap_reporter.py	Deterministic gap_id; fix grouping refs; add 409 reopen retry + error logging.
redgnat/engagement/token.py	Switch to datetime.UTC usage.
redgnat/engagement/kill_switch.py	Fail-closed kill switch design; durable-store unreachable handling.
redgnat/engagement/gate.py	Gate 2 shared-secret match option; fail-closed authorize/check adjustments.
redgnat/emulation/tasks.py	Add probe-depth parsing + runaway guard; thread depth into triggered_by.
redgnat/emulation/runner.py	Safety check before each step; status aggregation; bounded rate pause; fail-closed on safety check errors.
redgnat/emulation/plan.py	Remove circular type import; keep technique_cls generic.
redgnat/config.py	Add phase2_unlock_secret, max_inline_poll_seconds, feedback_max_probe_depth.
redgnat/client.py	Remove unused builder import; intake path cleanup.
redgnat/api/routes/stix.py	UTC handling fixes for STIX timestamps.
redgnat/api/routes/scenarios.py	Chain HTTPException cause for ValueError->404 mapping.
redgnat/api/routes/intel.py	Import ordering cleanup.
redgnat/api/routes/engage.py	Require X-Kill-Key for authorize; duration cap; improved error mapping.
redgnat/api/app.py	Expand CORS to DELETE + X-Kill-Key; router import order cleanup.
pyproject.toml	Ruff config: allow FastAPI DI defaults for bugbear.
config/config.ini.example	Document new config keys (GNAT API base, unlock secret, poll cap, probe depth).

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

+        entra_domain = cfg.entra_tenant_id if "." in (cfg.entra_tenant_id or "") else ""
+        okta_host = (cfg.okta_base_url or "").split("://")[-1].split("/")[0]
+
+        if "entra" in providers and cfg.entra_tenant_id and not _domain_allowed(entra_domain):


+        okta_host = (cfg.okta_base_url or "").split("://")[-1].split("/")[0]

-        if "entra" in providers and cfg.entra_tenant_id:
+        if "entra" in providers and cfg.entra_tenant_id and not _domain_allowed(entra_domain):


+    def test_in_scope_tenant_is_enumerated(self):
+        t = CloudEnumTechnique()
+        with patch("redgnat.config.RedGNATConfig", return_value=self._cfg()), \
+             patch.object(t, "_enum_entra", return_value=[{"user": "a"}]) as entra, \
+             patch.object(t, "_enum_okta", return_value=[]):
+            result = t.execute(_ctx(domains=["corp.example.com"]))
+        entra.assert_called_once()
+        assert result.status == ResultStatus.SUCCESS
+
+


wrhalpin · 2026-07-05T00:18:50Z

Good catch on the Entra scope gate — that was a real regression. A GUID entra_tenant_id (the common case, as in config.ini.example) produced an empty entra_domain, and _domain_allowed("") returned False whenever target_domains was configured, hard-blocking all Entra enumeration/analysis.

Fixed in 903bb7b: the domain gate now only applies to a domain-form tenant and is skipped for GUID tenants (in both cloud_enum.py and token_theft.py). Added test_guid_tenant_is_not_domain_gated regression tests for both.

The same commit also gets CI green: corrected the PEP 517 build backend (setuptools.backends.legacy:build didn't exist → setuptools.build_meta, which was failing pip install -e and masking the test/mypy/hygiene jobs), and cleared the repo-wide ruff check/ruff format and mypy redgnat/ failures.

Generated by Claude Code

Safety controls previously failed open on infrastructure faults: - KillSwitch.is_active() only consulted the durable Postgres record when Redis raised; a normal Redis cache-miss (eviction/restart) returned False even while a durable kill record stood. Redis is now only a fast path for the positive answer; Postgres is authoritative for "clear" and the check fails closed (halts) when the durable store is unreachable. - The runner checked the kill switch only *between* steps and discarded the result for single-step (probe) runs, so the first technique of any run always fired. Safety is now checked before every step, including the first and single-step plans. - Runner/EngagementRunner swallowed kill/gate-check exceptions and continued; they now halt on error. - Added RunStatus.EXPIRED and run-status aggregation so token-expiry halts are distinct from kills and an all-error run is FAILED, not COMPLETED. https://claude.ai/code/session_01DGwbYiq7PnxMxMXtnTEmov Signed-off-by: Claude <noreply@anthropic.com>

- POST /engage/authorize now requires the privileged X-Kill-Key header (the same elevated key that guards the kill switch) and is denied when that key is unconfigured, so a holder of only the general API key can no longer mint the Gate-3 engagement token. Revoke already required it; minting was the weaker of the two. - The route now caps duration at 24h and maps gate errors correctly: RuntimeError (Gate 1/2 unsatisfied) -> 403, ValueError -> 422, instead of leaking an opaque 500. - Gate 2 can now verify REDGNAT_PHASE2_UNLOCK against a configured phase2_unlock_secret (a real shared secret) instead of presence only; presence-only remains the default when no secret is configured. https://claude.ai/code/session_01DGwbYiq7PnxMxMXtnTEmov Signed-off-by: Claude <noreply@anthropic.com>

- Scope.allows_cidr now requires full containment (subnet_of a target range) instead of mere overlap, so a broad CIDR (10.0.0.0/8) can no longer pass validation against a narrow target (10.50.0.0/16) and then be scanned in full. - ad_enum takes its LDAP server and bind credentials from trusted config only (never ctx.params) and validates the server host against scope before binding, closing a credential-exfiltration path where an intel-driven param could redirect the bind to an attacker host. - cloud_enum and token_theft now enforce domain scope on the Entra/Okta tenant when target_domains is configured. - password_spray and credential_stuffing no longer divide by max_rate_per_minute unguarded (ZeroDivisionError when set to 0). - oauth_abuse target-domain extraction was always yielding "" (technique was permanently dead); fixed to rsplit on "@". - The four campaign-result polls now cap the inline wait at gophish.max_inline_poll_seconds so a handful of phishing runs cannot starve the Celery worker pool; the campaign keeps running in GoPhish. - Phishing/oauth techniques now tear down created GoPhish resources (complete campaign + delete page/template/group) on error instead of leaving a live campaign sending. - spearphishing_link now sets send_by_date from campaign_hours (was ""). https://claude.ai/code/session_01DGwbYiq7PnxMxMXtnTEmov Signed-off-by: Claude <noreply@anthropic.com>

- Sighting.sighting_of_ref was "attack-pattern--T1046" — not a valid STIX 2.1 id (type--UUID). It now derives a stable UUIDv5 from the ATT&CK id (preserved in x_redgnat_metadata.attack_technique_id) so GNAT can ingest the sighting. - gap_id is now deterministic (uuid5 of the run_id) so the note id emitted by /stix/gaps matches the note--<gap_id> reference in the run's Grouping and connector re-pulls don't create duplicate Notes. - Intake now assigns deterministic feed_id (uuid5 of source+ref) and the normalizer a deterministic scenario_id (uuid5 of the feed), so re-polling the same GNAT campaign / SandGNAT analysis upserts the same rows instead of accumulating a new feed+scenario+run every poll cycle. - Added a probe-depth runaway guard: ProbeRequest carries a generation depth, probe-driven runs are tagged probe:<id>:d<N>, and feedback stops generating probes once feedback.max_probe_depth (default 3) is reached — bounding the self-amplifying gap->probe->emulate loop whose rule table otherwise cycles forever. - The investigation evidence push now honors the returned error, retries with reopen on a 409 (closed investigation), and logs a real failure instead of silently reporting success; the pushed Grouping no longer references a CoA object absent from the bundle. - RedGNATConnector.list_objects now supports "grouping" (was silently []). https://claude.ai/code/session_01DGwbYiq7PnxMxMXtnTEmov Signed-off-by: Claude <noreply@anthropic.com>

- ScenarioBuilder now resolves tactic/name via the parent-fallback mapper lookups instead of raw get(), so a subtechnique missing from the static map no longer records tactic="unknown"; technique_name gained the same parent fallback technique_tactic already had. - token_theft: the dead mfa_bypass computation now produces a real finding, and both impossible-travel detectors apply the _IMPOSSIBLE_TRAVEL_MINUTES time window instead of flagging every different-IP pair. - CORS now allows DELETE and the X-Kill-Key header so a browser console at a configured origin can reach the kill-switch/revoke endpoints. - Removed the dead ScenarioBuilder instantiation in client.ingest_latest. - Clarified the synthetic T1046.service registry alias. - Lint: applied safe ruff autofixes (unused imports, import sorting, datetime UTC), added raise-from at three sites, modernized the registry type annotations, and configured the FastAPI DI idiom exception (92 -> 37 lint findings; remainder are deliberate quoted forward-refs and behavioral str-enum changes left as-is). https://claude.ai/code/session_01DGwbYiq7PnxMxMXtnTEmov Signed-off-by: Claude <noreply@anthropic.com>

Adds phase2_unlock_secret, gnat.api_base_url/api_key, gophish.max_inline_poll_seconds, and feedback.max_probe_depth introduced by the bug-fix pass. https://claude.ai/code/session_01DGwbYiq7PnxMxMXtnTEmov Signed-off-by: Claude <noreply@anthropic.com>

Adds unit tests for the previously-untested layers, mocking each module's external boundary (DB, HTTP, Celery, GNAT/GoPhish clients): - API layer via FastAPI TestClient — every scenarios/runs/intel/stix/engage route, auth, error paths, and CART report rendering (was 0%). - ScenarioStore with a mocked psycopg connection — all upsert/get/list methods and the positional row mappers. - EmulationRunner / EngagementRunner — success, all-error->FAILED, kill-before-first-step, empty plan, fail-closed on check error, and token-expiry->EXPIRED. - Celery task bodies — run_scenario/run_probe/run_engagement/ingest tasks, the feedback loop, and the probe-depth runaway guard. - RedGNATClient facade, both intake subscribers (poll/dedup/skip paths), every RedGNATConfig property, and the GoPhish-backed technique success and teardown-on-error paths. 288 tests pass; coverage 70.64% so `make coverage` now passes its gate. https://claude.ai/code/session_01DGwbYiq7PnxMxMXtnTEmov Signed-off-by: Claude <noreply@anthropic.com>

Targets the decision points that must be tested, not raw coverage: - Credential attacks (Control #4): password_spray/credential_stuffing only act on scope.target_accounts and reject attacker-supplied out-of-scope usernames; mfa_fatigue enforces the confirm gate, the push-count cap, and the password/account requirements; the max_rate_per_minute=0 path no longer risks ZeroDivisionError. - Scope gates added in the fix pass: ad_enum refuses an out-of-scope LDAP host and ignores a params-supplied server (credential-exfil guard); cloud_enum and token_theft skip out-of-scope Entra/Okta tenants. - gap_reporter push internals: the evidence-endpoint push, the 409->reopen retry, terminal-failure handling, and the GNATClient fallback (success, not-installed, and error branches). - EmulationRunner/EngagementRunner fail-closed: the engagement gate raising (not just denying) halts the run, and a store error mid-run marks it FAILED. Coverage 70.6% -> 76%. Credential techniques 30-40% -> 54-80%, ad_enum 26% -> 52%, gap_reporter 67% -> 82%, runner 88% -> 94%. The lines still uncovered are the HTTP/LDAP enumeration mechanics (network wrappers), not safety controls. 316 tests pass. https://claude.ai/code/session_01DGwbYiq7PnxMxMXtnTEmov Signed-off-by: Claude <noreply@anthropic.com>

- Correct the PEP 517 build backend (setuptools.backends.legacy:build did not exist -> setuptools.build_meta), which was failing pip install -e in CI and masking the test/mypy/hygiene jobs. - Fix the Entra scope-gate regression flagged in review: a GUID tenant_id (the common case) yielded an empty domain and was force-blocked; the gate now only applies to a domain-form tenant and is skipped for GUIDs, in both cloud_enum and token_theft. Added GUID-tenant regression tests. - Make `ruff check .` and `ruff format --check .` pass repo-wide: resolve quoted forward-refs via TYPE_CHECKING imports, combine nested with- statements, drop unused vars, ignore UP042 (the dataclass-ORM enums keep (str, Enum) deliberately), and apply the formatter. - Make `mypy redgnat/` pass: type the runner's scenario/step params, simplify the password-spray client dispatch, guard the SandGNAT bundle type, annotate a few locals, and remove redundant type: ignore comments. https://claude.ai/code/session_01DGwbYiq7PnxMxMXtnTEmov Signed-off-by: Claude <noreply@anthropic.com>

The GNAT platform library is not published to public PyPI (only an unrelated `gnat` 0.x exists there), so declaring `gnat>=1.5.0` as a hard dependency made `pip install` fail in CI and any environment without a private index. Every import of it in this package is already lazy and guarded, and the tests mock it, so it belongs in an optional `[gnat]` extra (install `redgnat[gnat]` or provide it via a private index). https://claude.ai/code/session_01DGwbYiq7PnxMxMXtnTEmov Signed-off-by: Claude <noreply@anthropic.com>

@master

- Add a minimal pylint config: fail-under = 8.5 (current score 8.71) and ignored-modules for the optional/lazy-imported integrations (gnat, ldap3, nmap, boto3) so their absence in lint envs isn't reported as import-error. The codebase deliberately uses lazy imports and shares result-shaping across parallel technique modules, so quality is enforced via a score floor rather than a perfect-10 chase. - Pin the DCO action to tim-actions/dco@v1.1.0 and give checkout fetch-depth: 0 so it can read the full PR commit range (the @master + shallow-clone combination failed with "Unexpected end of JSON input"). https://claude.ai/code/session_01DGwbYiq7PnxMxMXtnTEmov Signed-off-by: Claude <noreply@anthropic.com>

tim-actions/dco fails on current GitHub runners regardless of version or checkout depth ("Unexpected end of JSON input" — it's built for Node 20 and its JSON parsing breaks under Node 24). Replace it with an equivalent inline check that validates every commit in the PR range carries a Signed-off-by trailer. All existing commits are already signed off, so this passes. https://claude.ai/code/session_01DGwbYiq7PnxMxMXtnTEmov Signed-off-by: Claude <noreply@anthropic.com>

Copilot AI review requested due to automatic review settings July 4, 2026 23:38

Copilot started reviewing on behalf of wrhalpin July 4, 2026 23:39 View session

Copilot AI reviewed Jul 4, 2026

View reviewed changes

claude added 10 commits July 5, 2026 08:27

wrhalpin force-pushed the claude/refactor-pages-docs-fTvy5 branch from 903bb7b to fa3617e Compare July 5, 2026 08:27

claude added 2 commits July 5, 2026 08:35

wrhalpin merged commit 8f00250 into main Jul 5, 2026
9 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix safety-critical bugs and raise test coverage to 76% - #13

Fix safety-critical bugs and raise test coverage to 76%#13
wrhalpin merged 12 commits into
mainfrom
claude/refactor-pages-docs-fTvy5

wrhalpin commented Jul 4, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

wrhalpin commented Jul 5, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

wrhalpin commented Jul 4, 2026

Summary

Bug fixes

Testing

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

wrhalpin commented Jul 5, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants