fix(agents): preserve format guard recovery

.github/scripts/github-api-with-retry.js

@@ -170,7 +170,7 @@ function logWithCore(core, level, message) {
     core[level](message);
     return;
   }
-  const logFn = level === 'error' ? console.error : level === 'warning' ? console.warn : console.error;
+  const logFn = level === 'error' ? console.error : level === 'warning' ? console.warn : console.log;
   logFn(message);
 }
 

.github/workflows/agents-issue-format-guard.yml

@@ -202,8 +202,34 @@ jobs:
           # GITHUB_TOKEN label edits do not start issues:labeled workflows; dispatch is explicit.
           # Persist the completion marker only after a successful workflow_dispatch so a
           # failed run remains retryable on the next guard pass.
+          # Capture the attempt clock before dispatch: gh workflow run can exit non-zero
+          # after GitHub has already accepted the request.
+          dispatch_attempted_at=$(date -u +%Y-%m-%dT%H:%M:%SZ)
           if ! gh workflow run agents-issue-optimizer.yml --repo "$GITHUB_REPOSITORY" \
               -f issue_number="$NUMBER" -f phase=format; then
+            accepted=false
+            for _try in 1 2 3 4 5; do
+              sleep 2
+              # shellcheck disable=SC2016
+              match_count=$(gh run list --repo "$GITHUB_REPOSITORY" \
+                --workflow=agents-issue-optimizer.yml \
+                --event workflow_dispatch \
+                --limit 20 \
+                --json createdAt,displayTitle \
+                | jq --arg since "$dispatch_attempted_at" --arg issue "#$NUMBER" \
+                  '[.[] | select(.createdAt >= $since and (.displayTitle | contains($issue)))] | length')
+              if [[ "${match_count:-0}" -gt 0 ]]; then
+                accepted=true
+                break
+              fi
+            done
+            if [[ "$accepted" == true ]]; then
+              echo "::warning::optimizer dispatch CLI failed but a matching run was accepted; preserving agents:format lease"
+              echo "::error::optimizer dispatch status ambiguous; no completion marker written"
+              exit 1
+            fi
+            gh issue edit "$NUMBER" --repo "$GITHUB_REPOSITORY" --remove-label "agents:format" \
+              || echo "::warning::could not release failed agents:format lease"
             echo "::error::optimizer dispatch failed; no completion marker written so later runs can retry"
             exit 1
           fi

.github/workflows/agents-issue-optimizer.yml

@@ -349,6 +349,7 @@ jobs:
           });
           NODE
 
+          set +e
           python - <<'PY'
           import json
           from scripts.langchain import issue_dedup
@@ -385,6 +386,7 @@ jobs:
                   out.write(comment)
           PY
           dedup_rc=$?
+          set -e
           if [[ $dedup_rc -ne 0 ]]; then
             echo "::warning::issue dedup python exited with $dedup_rc; continuing without similar-issues comment"
           fi

config/template-drift-allowlist.txt

@@ -113,9 +113,9 @@ reason = Intentional divergence re-baselined 2026-06-30: root and consumer guard
 [pair.11]
 main = .github/workflows/agents-issue-optimizer.yml
 template = templates/consumer-repo/.github/workflows/agents-issue-optimizer.yml
-main_sha256 = ccf1cd3ccdba84c54743d2950367308df15c5928b95c1e7400a5d2858b7c1fa5
-template_sha256 = a4d50beb165724ff9266a18390d738105538fe0c13a5eb3a620fb8c9297fe80f
-reason = Intentional divergence re-baselined 2026-08-08b: root remains in-tree (scripts/langchain + .github/scripts/issue_format.py); consumer vendors those via Workflows sparse-checkout under workflows-scripts/. Shared behavioral contract this round: concurrency includes inputs.issue_number for workflow_dispatch dedupe, cancel-safe (failure()||cancelled()) agents:format lease release, and fixed issue_dedup Python indentation with visible non-zero exit warnings. Do not align wholesale — that would strip consumer action pins/token setup.
+main_sha256 = 128b60e2af8e3957209849c7f20502a1b9a971e0c8c347e25a0d501a868bc484
+template_sha256 = f7987a566c15971ae9039bb3d1a53f0bb654378e6c1dc3103f5692f5abca022d
+reason = Intentional divergence re-baselined 2026-08-08c: root remains in-tree (scripts/langchain + .github/scripts/issue_format.py); consumer vendors those via Workflows sparse-checkout under workflows-scripts/. Shared behavioral contract this round: concurrency includes inputs.issue_number for workflow_dispatch dedupe, cancel-safe (failure()||cancelled()) agents:format lease release, issue_dedup wrapped in set +e/set -e so advisory failures cannot abort under bash -e, and visible non-zero exit warnings. Do not align wholesale — that would strip consumer action pins/token setup.
 
 [pair.12]
 main = .github/workflows/agents-keepalive-loop-reporter.yml

langsmith-fleet-worker-attempt.json

@@ -1,13 +1,13 @@
 {
   "agent": "codex",
   "cli_version": "0.144.1",
-  "emitted_at": "2026-08-08T11:32:33.483036Z",
+  "emitted_at": "2026-08-08T15:51:23.873676Z",
   "execution_profile": "codex-default",
   "fallback_models": [
     "gpt-5.5"
   ],
   "operation_role": "worker",
-  "pr_number": "2981",
+  "pr_number": "2985",
   "requested_model": "gpt-5.6-terra",
   "runner": "reusable-codex-run",
   "schema": "langsmith-fleet/v1",

templates/consumer-repo/.github/scripts/github-api-with-retry.js

@@ -170,7 +170,7 @@ function logWithCore(core, level, message) {
     core[level](message);
     return;
   }
-  const logFn = level === 'error' ? console.error : level === 'warning' ? console.warn : console.error;
+  const logFn = level === 'error' ? console.error : level === 'warning' ? console.warn : console.log;
   logFn(message);
 }
 

templates/consumer-repo/.github/workflows/agents-issue-format-guard.yml

@@ -202,8 +202,34 @@ jobs:
           # GITHUB_TOKEN label edits do not start issues:labeled workflows; dispatch is explicit.
           # Persist the completion marker only after a successful workflow_dispatch so a
           # failed run remains retryable on the next guard pass.
+          # Capture the attempt clock before dispatch: gh workflow run can exit non-zero
+          # after GitHub has already accepted the request.
+          dispatch_attempted_at=$(date -u +%Y-%m-%dT%H:%M:%SZ)
           if ! gh workflow run agents-issue-optimizer.yml --repo "$GITHUB_REPOSITORY" \
               -f issue_number="$NUMBER" -f phase=format; then
+            accepted=false
+            for _try in 1 2 3 4 5; do
+              sleep 2
+              # shellcheck disable=SC2016
+              match_count=$(gh run list --repo "$GITHUB_REPOSITORY" \
+                --workflow=agents-issue-optimizer.yml \
+                --event workflow_dispatch \
+                --limit 20 \
+                --json createdAt,displayTitle \
+                | jq --arg since "$dispatch_attempted_at" --arg issue "#$NUMBER" \
+                  '[.[] | select(.createdAt >= $since and (.displayTitle | contains($issue)))] | length')
+              if [[ "${match_count:-0}" -gt 0 ]]; then
+                accepted=true
+                break
+              fi
+            done
+            if [[ "$accepted" == true ]]; then
+              echo "::warning::optimizer dispatch CLI failed but a matching run was accepted; preserving agents:format lease"
+              echo "::error::optimizer dispatch status ambiguous; no completion marker written"
+              exit 1
+            fi
+            gh issue edit "$NUMBER" --repo "$GITHUB_REPOSITORY" --remove-label "agents:format" \
+              || echo "::warning::could not release failed agents:format lease"
             echo "::error::optimizer dispatch failed; no completion marker written so later runs can retry"
             exit 1
           fi

templates/consumer-repo/.github/workflows/agents-issue-optimizer.yml

@@ -348,6 +348,7 @@ jobs:
           });
           NODE
 
+          set +e
           python - <<'PY'
           import json
           import sys
@@ -386,6 +387,7 @@ jobs:
                   out.write(comment)
           PY
           dedup_rc=$?
+          set -e
           if [[ $dedup_rc -ne 0 ]]; then
             echo "::warning::issue dedup python exited with $dedup_rc; continuing without similar-issues comment"
           fi

tests/workflows/test_agents_issue_optimizer_format_trigger.py

@@ -81,6 +81,16 @@ def test_format_lease_is_required_and_released_after_failure() -> None:
         assert "exit 1" in text[lease_idx:dispatch_idx]
         # Dispatch must stay inside the success path after lease acquisition.
         assert '--add-label "agents:format"' in text[:dispatch_idx]
+        # Clock capture precedes the dispatch so an accepted-but-errored CLI can be
+        # reconciled; lease release stays inside the failure block only after that probe.
+        failure_start = text.index("if ! gh workflow run agents-issue-optimizer.yml")
+        assert "dispatch_attempted_at=" in text[dispatch_idx - 200 : failure_start]
+        failure_end = text.index("\n          fi", failure_start)
+        failure_block = text[failure_start:failure_end]
+        assert "preserving agents:format lease" in failure_block
+        assert '--remove-label "agents:format"' in failure_block
+        assert '|| echo "::warning::could not release failed agents:format lease"' in failure_block
+        assert "no completion marker written so later runs can retry" in failure_block
 
     for text in (
         WORKFLOW_PATH.read_text(encoding="utf-8"),