[codex] Recent unresolved review feedback

[haasonsaas]

Recent unresolved review feedback

• Owner: evalops
• Merged since: 2026-05-13
• Minimum severity: high
• Findings: 8

• p1 evalops/deploy#3061 scripts/reconcile-bazel-buildfarm-ops-agent.sh:63
  • PR: https://github.com/evalops/deploy/pull/3061
  • Feedback: https://github.com/evalops/deploy/pull/3061#discussion_r3252557377
  • Propagate base64 decode failures before systemd-run
• p1 evalops/deploy#3048 scripts/github-actions-ci-metrics.py:497
  • PR: https://github.com/evalops/deploy/pull/3048
  • Feedback: https://github.com/evalops/deploy/pull/3048#discussion_r3252431806
  • Evaluate buildfarm pressure from completed jobs only
• high evalops/nimbus#40 scripts/run-bazel-rbe.sh:28
  • PR: https://github.com/evalops/nimbus/pull/40
  • Feedback: https://github.com/evalops/nimbus/pull/40#discussion_r3252405608

  • Env var BAZEL_RBE_USE_EXISTING_TUNNEL set but never checked

• high evalops/console#42 scripts/run-bazel-rbe.sh:27
  • PR: https://github.com/evalops/console/pull/42
  • Feedback: https://github.com/evalops/console/pull/42#discussion_r3252408138

  • Env var BAZEL_RBE_USE_EXISTING_TUNNEL set but never read

• high [codex] add Bazel RBE smoke kestrel#99 Makefile:26
  • PR: [codex] add Bazel RBE smoke kestrel#99
  • Feedback: [codex] add Bazel RBE smoke kestrel#99 (comment)

  • Script ignores BAZEL_TARGETS, always runs all tests

• p1 evalops/maestro-internal#1976 pr_review
  • PR: https://github.com/evalops/maestro-internal/pull/1976
  • Feedback: https://github.com/evalops/maestro-internal/pull/1976#pullrequestreview-4303199202
  • Continue refreshing interrupted A2A task states
• high ci: add Bazel RBE lane mocktopus#9 .github/workflows/bazel-rbe.yml:66
  • PR: ci: add Bazel RBE lane mocktopus#9
  • Feedback: ci: add Bazel RBE lane mocktopus#9 (comment)

  • CI workflow missing uv needed by bazel-check

• p1 fix: omit non-terminal runtime promotion writes maestro#421 src/server/agent-runtime-ledger.ts:293
  • PR: fix: omit non-terminal runtime promotion writes maestro#421
  • Feedback: fix: omit non-terminal runtime promotion writes maestro#421 (comment)
  • Preserve wait_run emission for pending governance entries

[haasonsaas]

Recent unresolved review feedback

• Owner: evalops
• Merged since: 2026-05-13
• Minimum severity: high
• Findings: 5

• p1 evalops/deploy#3129 contracts/release-trains/v1/trains.yaml:43
  • PR: https://github.com/evalops/deploy/pull/3129
  • Feedback: https://github.com/evalops/deploy/pull/3129#discussion_r3253249768
  • Align new train promotion workflow with train ID wiring
• p1 evalops/platform#2103 internal/agentruntime/agentruntime/operating_plane.go:239
  • PR: https://github.com/evalops/platform/pull/2103
  • Feedback: https://github.com/evalops/platform/pull/2103#discussion_r3253120972
  • Filter work items by audience before projecting them
• p1 evalops/deploy#3103 k8s/production/jobs/tool-execution-iam-grants.yaml:107
  • PR: https://github.com/evalops/deploy/pull/3103
  • Feedback: https://github.com/evalops/deploy/pull/3103#discussion_r3253059636
  • Create IAM DB user via Cloud SQL API, not CREATE ROLE
• p1 evalops/platform#2100 internal/agentruntime/a2a/http.go:298
  • PR: https://github.com/evalops/platform/pull/2100
  • Feedback: https://github.com/evalops/platform/pull/2100#discussion_r3252951466
  • Avoid truncating work-graph metadata to 100 work items
• p1 evalops/platform#2098 internal/agentruntime/a2a/http.go:238
  • PR: https://github.com/evalops/platform/pull/2098
  • Feedback: https://github.com/evalops/platform/pull/2098#discussion_r3252932732
  • Preserve idempotent retries for resumed task replies

[haasonsaas]

Recent unresolved review feedback

• Owner: evalops
• Merged since: 2026-05-14
• Minimum severity: high
• Findings: 1

• p1 evalops/maestro-internal#1989 src/platform/a2a-client.ts:625
  • PR: https://github.com/evalops/maestro-internal/pull/1989
  • Feedback: https://github.com/evalops/maestro-internal/pull/1989#discussion_r3253382729
  • Unwrap JSON-RPC stream envelopes before type inference

[haasonsaas]

Recent unresolved review feedback

• Owner: evalops
• Merged since: 2026-05-14
• Minimum severity: high
• Findings: 7

• p0 evalops/platform#2131 db/agentruntime/migrations/011_a2a_push_deliveries.up.sql:1
  • PR: https://github.com/evalops/platform/pull/2131
  • Feedback: https://github.com/evalops/platform/pull/2131#discussion_r3254155664
  • Restore migration SQL instead of snapshot pointer
• p0 evalops/platform#2131 internal/agentruntime/a2a/push.go:1
  • PR: https://github.com/evalops/platform/pull/2131
  • Feedback: https://github.com/evalops/platform/pull/2131#discussion_r3254155667
  • Revert local snapshot marker from Go source
• p1 evalops/platform#2125 internal/agentruntime/a2a/push.go:712
  • PR: https://github.com/evalops/platform/pull/2125
  • Feedback: https://github.com/evalops/platform/pull/2125#discussion_r3254087275
  • Canonicalize callback path before prefix allowlist checks
• p1 evalops/platform#2120 internal/agentruntime/a2a/push.go:306
  • PR: https://github.com/evalops/platform/pull/2120
  • Feedback: https://github.com/evalops/platform/pull/2120#discussion_r3254049708
  • Send final status only after terminal payloads
• p1 evalops/platform#2120 internal/agentruntime/a2a/push.go:291
  • PR: https://github.com/evalops/platform/pull/2120
  • Feedback: https://github.com/evalops/platform/pull/2120#discussion_r3254049709
  • Pin callback connection to validated IP
• p1 evalops/platform#2117 internal/codexworker/codex_workgraph.go:189
  • PR: https://github.com/evalops/platform/pull/2117
  • Feedback: https://github.com/evalops/platform/pull/2117#discussion_r3253962067
  • Keep heartbeats active while executing governed subagent calls
• p1 evalops/platform#2114 internal/codexworker/trace.go:453
  • PR: https://github.com/evalops/platform/pull/2114
  • Feedback: https://github.com/evalops/platform/pull/2114#discussion_r3253908766
  • Keep synthesized child_run_id within compact-id limit

[haasonsaas]

Recent unresolved review feedback

• Owner: evalops
• Merged since: 2026-05-14
• Minimum severity: high
• Findings: 4

• p1 evalops/platform#2178 internal/agentruntime/agentruntime/operating_plane.go:795
  • PR: https://github.com/evalops/platform/pull/2178
  • Feedback: https://github.com/evalops/platform/pull/2178#discussion_r3254632131
  • Align session lookup with newly accepted session ID fields
• p1 evalops/platform#2163 internal/agentregistry/http/router.go:375
  • PR: https://github.com/evalops/platform/pull/2163
  • Feedback: https://github.com/evalops/platform/pull/2163#discussion_r3254470598
  • Gate A2A dispatcher on non-empty endpoint allowlist
• p1 evalops/platform#2150 internal/agentregistry/agents/redis_delegations.go:62
  • PR: https://github.com/evalops/platform/pull/2150
  • Feedback: https://github.com/evalops/platform/pull/2150#discussion_r3254346340
  • Avoid treating terminal replay resolves as state changes
• p1 evalops/deploy#3329 scripts/check-agent-runtime-a2a-smoke.py:1247
  • PR: https://github.com/evalops/deploy/pull/3329
  • Feedback: https://github.com/evalops/deploy/pull/3329#discussion_r3254365120
  • Avoid single-attempt work-graph fetch in default smoke mode

[haasonsaas]

Recent unresolved review feedback

• Owner: evalops
• Merged since: 2026-05-14
• Minimum severity: high
• Findings: 6

• p1 evalops/platform#2210 internal/agentruntime/agentruntime/operating_plane.go:1438
  • PR: https://github.com/evalops/platform/pull/2210
  • Feedback: https://github.com/evalops/platform/pull/2210#discussion_r3254864417
  • Preserve global cost keys when session id is set
• p1 evalops/platform#2206 internal/agentruntime/agentruntime/run_operations.go:742
  • PR: https://github.com/evalops/platform/pull/2206
  • Feedback: https://github.com/evalops/platform/pull/2206#discussion_r3254880784
  • Page through all pending timer waits before scheduling wake
• p1 evalops/deploy#3406 environments/production/platform-runtime.lock.yaml:89
  • PR: https://github.com/evalops/deploy/pull/3406
  • Feedback: https://github.com/evalops/deploy/pull/3406#discussion_r3254827177
  • Pin remote-runner to receipt-backed digest
• p1 evalops/platform#2197 internal/agentregistry/agents/redis_delegations.go:44
  • PR: https://github.com/evalops/platform/pull/2197
  • Feedback: https://github.com/evalops/platform/pull/2197#discussion_r3254753836
  • Skip Redis capacity CAS when target has unlimited capacity
• p1 evalops/platform#2197 internal/agentregistry/agents/redis_delegations.go:235
  • PR: https://github.com/evalops/platform/pull/2197
  • Feedback: https://github.com/evalops/platform/pull/2197#discussion_r3254753837
  • Avoid writing watched agent state when no slot was reserved
• p1 evalops/deploy#3399 scripts/check-platform-runtime-rollouts.py:967
  • PR: https://github.com/evalops/deploy/pull/3399
  • Feedback: https://github.com/evalops/deploy/pull/3399#discussion_r3254781571
  • Evaluate full pod selectors before accepting policy routes

[haasonsaas]

Recent unresolved review feedback

• Owner: evalops
• Merged since: 2026-05-15
• Minimum severity: high
• Findings: 7

• p1 evalops/platform#2250 cmd/a2a-swarm-staging-smoke/main.go:1049
  • PR: https://github.com/evalops/platform/pull/2250
  • Feedback: https://github.com/evalops/platform/pull/2250#discussion_r3255774792
  • Preserve callback sequence when replaying dedupe checks
• p1 evalops/platform#2249 cmd/a2a-swarm-staging-smoke/main.go:784
  • PR: https://github.com/evalops/platform/pull/2249
  • Feedback: https://github.com/evalops/platform/pull/2249#discussion_r3255744815
  • Use CLI workflow-run fallback when building origin task URL
• p1 evalops/platform#2247 internal/agentruntime/agentruntime/operating_plane.go:2437
  • PR: https://github.com/evalops/platform/pull/2247
  • Feedback: https://github.com/evalops/platform/pull/2247#discussion_r3255693685
  • Redact push-delivery error strings before projecting
• p1 evalops/deploy#3496 scripts/check-agent-runtime-a2a-smoke.py:520
  • PR: https://github.com/evalops/deploy/pull/3496
  • Feedback: https://github.com/evalops/deploy/pull/3496#discussion_r3255523832
  • Skip delivered entries when checking current-task outbox failures
• p1 evalops/platform#2239 internal/agentruntime/agentruntime/platform_clients.go:107
  • PR: https://github.com/evalops/platform/pull/2239
  • Feedback: https://github.com/evalops/platform/pull/2239#discussion_r3255496059
  • Enforce run ownership on tool-execution read fallback
• p1 evalops/platform#2233 internal/agentruntime/a2a/http.go:2982
  • PR: https://github.com/evalops/platform/pull/2233
  • Feedback: https://github.com/evalops/platform/pull/2233#discussion_r3255448039
  • Apply skill governance before subagent request fields
• p1 evalops/deploy#3467 scripts/check-agent-runtime-a2a-smoke.py:1025
  • PR: https://github.com/evalops/deploy/pull/3467
  • Feedback: https://github.com/evalops/deploy/pull/3467#discussion_r3255336324
  • Redact operator markers before persisting diagnostics

[haasonsaas]

Recent unresolved review feedback

• Owner: evalops
• Merged since: 2026-05-15
• Minimum severity: high
• Findings: 6

• p1 evalops/deploy#3551 infrastructure/gcp/stacks/20-dns/tailscale-acl.tf:170
  • PR: https://github.com/evalops/deploy/pull/3551
  • Feedback: https://github.com/evalops/deploy/pull/3551#discussion_r3256330067
  • Preserve synthetic runner grants in managed ACL policy
• p1 evalops/platform#2263 internal/agentruntime/agentruntime/operating_plane.go:2619
  • PR: https://github.com/evalops/platform/pull/2263
  • Feedback: https://github.com/evalops/platform/pull/2263#discussion_r3256252598
  • Avoid seeding child run nodes with work-item state
• p1 evalops/deploy#3537 .github/workflows/a2a-swarm-staging-smoke.yml:266
  • PR: https://github.com/evalops/deploy/pull/3537
  • Feedback: https://github.com/evalops/deploy/pull/3537#discussion_r3256031576
  • Pass workspace ID when minting identity service token
• p1 evalops/platform#2250 cmd/a2a-swarm-staging-smoke/main.go:1049
  • PR: https://github.com/evalops/platform/pull/2250
  • Feedback: https://github.com/evalops/platform/pull/2250#discussion_r3255774792
  • Preserve callback sequence when replaying dedupe checks
• p1 evalops/platform#2249 cmd/a2a-swarm-staging-smoke/main.go:784
  • PR: https://github.com/evalops/platform/pull/2249
  • Feedback: https://github.com/evalops/platform/pull/2249#discussion_r3255744815
  • Use CLI workflow-run fallback when building origin task URL
• p1 evalops/platform#2247 internal/agentruntime/agentruntime/operating_plane.go:2437
  • PR: https://github.com/evalops/platform/pull/2247
  • Feedback: https://github.com/evalops/platform/pull/2247#discussion_r3255693685
  • Redact push-delivery error strings before projecting

[haasonsaas]

Recent unresolved review feedback

• Owner: evalops
• Merged since: 2026-05-15
• Minimum severity: high
• Findings: 7

• p1 evalops/platform#2283 internal/agentruntime/agentruntime/a2a_callback_artifacts.go:64
  • PR: https://github.com/evalops/platform/pull/2283
  • Feedback: https://github.com/evalops/platform/pull/2283#discussion_r3258811590
  • Re-emit artifact event on callback retry
• p1 evalops/platform#2281 internal/agentruntime/agentruntime/teammate_subagent_materialization.go:675
  • PR: https://github.com/evalops/platform/pull/2281
  • Feedback: https://github.com/evalops/platform/pull/2281#discussion_r3258718820
  • Preserve replayed child-run terminal state
• p1 evalops/platform#2280 internal/agentruntime/a2a/http.go:750
  • PR: https://github.com/evalops/platform/pull/2280
  • Feedback: https://github.com/evalops/platform/pull/2280#discussion_r3258595130
  • Handle missing delegation as a rejected callback
• p1 evalops/platform#2277 internal/agentregistry/agents/a2a_delegation_dispatcher.go:176
  • PR: https://github.com/evalops/platform/pull/2277
  • Feedback: https://github.com/evalops/platform/pull/2277#discussion_r3257846279
  • Fallback to static callback token when minting fails
• p1 evalops/deploy#3589 scripts/check-internal-runner-gcp-diagnostics.py:218
  • PR: https://github.com/evalops/deploy/pull/3589
  • Feedback: https://github.com/evalops/deploy/pull/3589#discussion_r3257563381
  • Measure queue age from job readiness, not run creation
• p1 evalops/deploy#3588 scripts/enforce-sync-staleness.py:165
  • PR: https://github.com/evalops/deploy/pull/3588
  • Feedback: https://github.com/evalops/deploy/pull/3588#discussion_r3257525286
  • Enable missing-manifest fail-fast for staleness lookups
• p1 evalops/deploy#3586 .github/workflows/sync-production-runtime-images.yml:1384
  • PR: https://github.com/evalops/deploy/pull/3586
  • Feedback: https://github.com/evalops/deploy/pull/3586#discussion_r3257443737
  • Skip branch commit when snapshot matches remote head

[haasonsaas]

Recent unresolved review feedback

• Owner: evalops
• Merged since: 2026-05-15
• Minimum severity: high
• Findings: 11

• p1 evalops/platform#2310 cmd/a2a-swarm-staging-smoke/main.go:4752
  • PR: https://github.com/evalops/platform/pull/2310
  • Feedback: https://github.com/evalops/platform/pull/2310#discussion_r3261203420
  • Use resumed task ID for a2a.resumes[0].run_id
• p1 evalops/deploy#3684 scripts/check-a2a-swarm-staging-gitops-convergence.py:383
  • PR: https://github.com/evalops/deploy/pull/3684
  • Feedback: https://github.com/evalops/deploy/pull/3684#discussion_r3260934001
  • Continue collecting results when a tracked workload is missing
• p1 evalops/platform#2308 internal/agentruntime/agentruntime/operating_plane.go:2525
  • PR: https://github.com/evalops/platform/pull/2308
  • Feedback: https://github.com/evalops/platform/pull/2308#discussion_r3260847412
  • Mark waits resolved when resolved_at is present
• p1 evalops/platform#2304 internal/agentruntime/agentruntime/run_operations.go:868
  • PR: https://github.com/evalops/platform/pull/2304
  • Feedback: https://github.com/evalops/platform/pull/2304#discussion_r3260506552
  • Preserve wait-group metadata when resuming grouped waits
• p0 evalops/platform#2302 cmd/a2a-swarm-staging-smoke/main.go:1
  • PR: https://github.com/evalops/platform/pull/2302
  • Feedback: https://github.com/evalops/platform/pull/2302#discussion_r3260706938
  • Restore buildable source for swarm smoke entrypoint
• p1 evalops/platform#2302 cmd/a2a-swarm-staging-smoke/main_test.go:1
  • PR: https://github.com/evalops/platform/pull/2302
  • Feedback: https://github.com/evalops/platform/pull/2302#discussion_r3260706942
  • Reinstate compilable test source for swarm smoke package
• p1 evalops/platform#2299 internal/agentregistry/agents/delegation.go:302
  • PR: https://github.com/evalops/platform/pull/2299
  • Feedback: https://github.com/evalops/platform/pull/2299#discussion_r3260284927
  • Parse all supported lane aliases for wait contracts
• p1 evalops/platform#2294 internal/agentruntime/a2a/http.go:4358
  • PR: https://github.com/evalops/platform/pull/2294
  • Feedback: https://github.com/evalops/platform/pull/2294#discussion_r3260101388
  • Prioritize callback correlation over operator snapshot fields
• p1 evalops/deploy#3645 .github/workflows/sync-production-runtime-images.yml:394
  • PR: https://github.com/evalops/deploy/pull/3645
  • Feedback: https://github.com/evalops/deploy/pull/3645#discussion_r3259618690
  • Exclude auth-style pull errors from missing-image classifier
• p1 evalops/deploy#3643 .github/workflows/sync-staging-runtime-images.yml:353
  • PR: https://github.com/evalops/deploy/pull/3643
  • Feedback: https://github.com/evalops/deploy/pull/3643#discussion_r3259581374
  • Fail the workflow on registry auth errors during mirroring
• p1 evalops/deploy#3643 pr_comment
  • PR: https://github.com/evalops/deploy/pull/3643
  • Feedback: https://github.com/evalops/deploy/pull/3643#issuecomment-4478583414
  • Fix-forward opened in #3645 for the late P1 review: auth/authorization pull errors now remain hard failures while true missing-image cases stay recoverable.

[haasonsaas]

Recent unresolved review feedback

• Owner: evalops
• Merged since: 2026-05-16
• Minimum severity: high
• Findings: 5

• p1 evalops/platform#2323 internal/agentruntime/a2a/http.go:1080
  • PR: https://github.com/evalops/platform/pull/2323
  • Feedback: https://github.com/evalops/platform/pull/2323#discussion_r3262806684
  • Broaden auth redaction to all supported secret keys
• p1 evalops/platform#2321 internal/agentruntime/a2a/http.go:1057
  • PR: https://github.com/evalops/platform/pull/2321
  • Feedback: https://github.com/evalops/platform/pull/2321#discussion_r3262795660
  • Do not return push callback secrets to task readers
• p1 evalops/platform#2315 internal/agentruntime/agentruntime/a2a_origin_run_materialization.go:503
  • PR: https://github.com/evalops/platform/pull/2315
  • Feedback: https://github.com/evalops/platform/pull/2315#discussion_r3262200629
  • Restrict swarm IDs to swarm metadata sources
• p1 evalops/deploy#3700 scripts/merge-ready-pr.py:255
  • PR: https://github.com/evalops/deploy/pull/3700
  • Feedback: https://github.com/evalops/deploy/pull/3700#discussion_r3261489523
  • Treat skipped required checks as passing
• p1 evalops/platform#2310 cmd/a2a-swarm-staging-smoke/main.go:4752
  • PR: https://github.com/evalops/platform/pull/2310
  • Feedback: https://github.com/evalops/platform/pull/2310#discussion_r3261203420
  • Use resumed task ID for a2a.resumes[0].run_id

[Ilya0527]

Two patterns worth flagging:

1. BAZEL_RBE_USE_EXISTING_TUNNEL appears twice — nimbus#40 line 28 and console#42 line 27, same script name (run-bazel-rbe.sh). That's one template-drift bug × 2 vendored copies, not two independent regressions. Does the sweep dedupe by (file_basename, finding_title_hash) across repos before ranking by severity? Otherwise a single shared-script defect inflates the count and obscures that fixing it once kills N findings. Same shape as BAZEL_TARGETS being ignored on kestrel#99 — knob advertised on the calling side, no reader in the script.

2. Set-but-never-read env vars are cheap to lint. Parse for ${VAR} reads vs VAR= exports, diff against the env-var surface documented in the calling workflow yaml. If a knob is advertised but no reader exists, fail CI. Three of the eight findings here are this same class.

Separate eye on the A2A item (maestro-internal#1976, "continue refreshing interrupted A2A task states"): interrupted-then-resumed task state is where orphan-tool-use hides. If the refresh loop doesn't observe the interruption boundary cleanly, the resumed task can re-emit a tool_use without an accompanying tool_result for the prior turn — looks like forward progress in the log, wedges silently downstream.

Diagnosis: ALEF-PAT-001 | Confidence: 0.5 | See standard: https://n50.io/patterns/001

[haasonsaas]

Recent unresolved review feedback

• Owner: evalops
• Merged since: 2026-05-16
• Minimum severity: high
• Findings: 3

• p1 evalops/platform#2323 internal/agentruntime/a2a/http.go:1080
  • PR: https://github.com/evalops/platform/pull/2323
  • Feedback: https://github.com/evalops/platform/pull/2323#discussion_r3262806684
  • Broaden auth redaction to all supported secret keys
• p1 evalops/platform#2321 internal/agentruntime/a2a/http.go:1057
  • PR: https://github.com/evalops/platform/pull/2321
  • Feedback: https://github.com/evalops/platform/pull/2321#discussion_r3262795660
  • Do not return push callback secrets to task readers
• p1 evalops/platform#2315 internal/agentruntime/agentruntime/a2a_origin_run_materialization.go:503
  • PR: https://github.com/evalops/platform/pull/2315
  • Feedback: https://github.com/evalops/platform/pull/2315#discussion_r3262200629
  • Restrict swarm IDs to swarm metadata sources

[haasonsaas]

Recent unresolved review feedback

• Owner: evalops
• Merged since: 2026-05-16
• Minimum severity: high
• Findings: 4

• p1 evalops/deploy#3771 k8s/production/agent-runtime/agent-runtime-deployment.yaml:177
  • PR: https://github.com/evalops/deploy/pull/3771
  • Feedback: https://github.com/evalops/deploy/pull/3771#discussion_r3265664038
  • Override the new durable in staging
• p1 evalops/platform#2344 internal/agentruntime/agentruntime/run_operations.go:830
  • PR: https://github.com/evalops/platform/pull/2344
  • Feedback: https://github.com/evalops/platform/pull/2344#discussion_r3265407857
  • Preserve wait contract before checking A2A completion
• p1 evalops/platform#2340 pkg/slacktracecontract/slack_replay.go:80
  • PR: https://github.com/evalops/platform/pull/2340
  • Feedback: https://github.com/evalops/platform/pull/2340#discussion_r3264566755
  • Reject documented raw_slack_text attributes
• p1 evalops/platform#2339 pr_comment
  • PR: https://github.com/evalops/platform/pull/2339
  • Feedback: https://github.com/evalops/platform/pull/2339#issuecomment-4485496477