From b8121fffd31e6415740d3f52d19e12defd9d28f9 Mon Sep 17 00:00:00 2001
From: stranske <stranske@gmail.com>
Date: Thu, 12 Feb 2026 00:59:57 +0000
Subject: [PATCH 01/11] fix: resolve 8 issues found in Codex run log audit

Essential fixes:
- Reporter sparse-checkout: add .github/actions to checkout so setup-api-client
  action is available (was failing 100% on Workflows repo)
- Belt Worker: re-install API client after branch checkout wipes node_modules
  (was causing @octokit/rest import failures and degraded token rotation)

High-value fixes:
- LLM analysis outputs: use print(..., end='') to strip trailing newlines from
  python extraction (confidence values had '\n' suffix e.g. '0.63\n')
- Repo variables fetch: downgrade from core.info to core.debug since the token
  permission limitation is known and the fallback to defaults works correctly

Medium fixes:
- Health 75 API Rate Diagnostic: pass secrets to 4 setup-api-client calls that
  were missing the input, causing 'No tokens were exported' warnings
- datetime.utcnow(): replace deprecated calls with timezone-aware alternative
  in both Belt Worker ledger functions

Low-salience fixes:
- error_classifier: gate entry log behind RUNNER_DEBUG to reduce log noise
- Non-artifact commit warning: downgrade from warning to notice since it is
  expected behavior when Codex produces only workflow artifacts
---
 .github/scripts/error_classifier.js           |  6 ++++--
 .github/scripts/keepalive_loop.js             |  2 +-
 .../workflows/agents-72-codex-belt-worker.yml | 11 ++++++++--
 .../agents-keepalive-loop-reporter.yml        |  1 +
 .../health-75-api-rate-diagnostic.yml         | 12 +++++++++++
 .github/workflows/reusable-codex-run.yml      | 20 +++++++++----------
 .../.github/scripts/error_classifier.js       |  6 ++++--
 .../.github/scripts/keepalive_loop.js         |  2 +-
 .../workflows/agents-72-codex-belt-worker.yml | 11 ++++++++--
 9 files changed, 51 insertions(+), 20 deletions(-)

diff --git a/.github/scripts/error_classifier.js b/.github/scripts/error_classifier.js
index 18f89c2c3..85709c43f 100644
--- a/.github/scripts/error_classifier.js
+++ b/.github/scripts/error_classifier.js
@@ -189,8 +189,10 @@ function classifyByMessage(message) {
 function classifyError(error) {
   const message = normaliseMessage(error);
   const preview = message ? message.slice(0, 50) : 'unknown';
-  // eslint-disable-next-line no-console
-  console.log(`[error_classifier] Classifying error: ${preview}`);
+  if (process.env.RUNNER_DEBUG === '1') {
+    // eslint-disable-next-line no-console
+    console.log(`[error_classifier] Classifying error: ${preview}`);
+  }
   const status = getStatusCode(error);
 
   const statusCategory = status ? classifyByStatus(status, message) : null;
diff --git a/.github/scripts/keepalive_loop.js b/.github/scripts/keepalive_loop.js
index 26938a680..96cdf80e4 100644
--- a/.github/scripts/keepalive_loop.js
+++ b/.github/scripts/keepalive_loop.js
@@ -374,7 +374,7 @@ async function fetchRepoVariables({ github, context, core, names = [] }) {
     }
   } catch (error) {
     if (core) {
-      core.info(`Failed to fetch repository variables for timeout config: ${error.message}`);
+      core.debug(`Repository variables not accessible for timeout config (using defaults): ${error.message}`);
     }
   }
 
diff --git a/.github/workflows/agents-72-codex-belt-worker.yml b/.github/workflows/agents-72-codex-belt-worker.yml
index 28dc732b5..95530873d 100644
--- a/.github/workflows/agents-72-codex-belt-worker.yml
+++ b/.github/workflows/agents-72-codex-belt-worker.yml
@@ -591,6 +591,13 @@ jobs:
           fetch-depth: 1
           path: .belt-tools
 
+      - name: Re-install API client after branch checkout
+        if: ${{ steps.parallel.outputs.allowed == 'true' && (inputs.keepalive != true || steps.keepalive_gate.outputs.action != 'skip') }}
+        uses: ./.github/actions/setup-api-client
+        with:
+          secrets: ${{ toJSON(secrets) }}
+          github_token: ${{ github.token }}
+
       - name: Validate ledger base branch
         if: ${{ steps.parallel.outputs.allowed == 'true' && (inputs.keepalive != true || steps.keepalive_gate.outputs.action != 'skip') }}
         env:
@@ -764,7 +771,7 @@ jobs:
                   return super().increase_indent(flow, False)
 
           def iso_now() -> str:
-              return dt.datetime.utcnow().replace(microsecond=0).isoformat() + 'Z'
+              return dt.datetime.now(dt.timezone.utc).replace(microsecond=0).isoformat().replace('+00:00', 'Z')
 
           issue = os.environ["ISSUE"].strip()
           branch = os.environ["BRANCH"].strip()
@@ -1272,7 +1279,7 @@ jobs:
                   return super().increase_indent(flow, False)
 
           def iso_now() -> str:
-              return dt.datetime.utcnow().replace(microsecond=0).isoformat() + 'Z'
+              return dt.datetime.now(dt.timezone.utc).replace(microsecond=0).isoformat().replace('+00:00', 'Z')
 
           issue = os.environ.get('ISSUE', '').strip()
           branch = os.environ.get('BRANCH', '').strip()
diff --git a/.github/workflows/agents-keepalive-loop-reporter.yml b/.github/workflows/agents-keepalive-loop-reporter.yml
index feeae1868..f9c8890d0 100644
--- a/.github/workflows/agents-keepalive-loop-reporter.yml
+++ b/.github/workflows/agents-keepalive-loop-reporter.yml
@@ -40,6 +40,7 @@ jobs:
           token: ${{ steps.app_token.outputs.token || github.token }}
           sparse-checkout: |
             .github/scripts
+            .github/actions
           sparse-checkout-cone-mode: false
           fetch-depth: 1
 
diff --git a/.github/workflows/health-75-api-rate-diagnostic.yml b/.github/workflows/health-75-api-rate-diagnostic.yml
index 0e28dae86..5dd070666 100644
--- a/.github/workflows/health-75-api-rate-diagnostic.yml
+++ b/.github/workflows/health-75-api-rate-diagnostic.yml
@@ -87,6 +87,9 @@ jobs:
         uses: actions/checkout@v6
       - name: Export load-balancer tokens
         uses: ./.github/actions/setup-api-client
+        with:
+          secrets: ${{ toJSON(secrets) }}
+          github_token: ${{ github.token }}
 
       - name: Check GITHUB_TOKEN rate limits
         id: github_token
@@ -1457,6 +1460,9 @@ jobs:
         uses: actions/checkout@v6
       - name: Export load-balancer tokens
         uses: ./.github/actions/setup-api-client
+        with:
+          secrets: ${{ toJSON(secrets) }}
+          github_token: ${{ github.token }}
       - name: Check for critical utilization
         id: check
         run: |
@@ -1570,6 +1576,9 @@ jobs:
         uses: actions/checkout@v6
       - name: Export load-balancer tokens
         uses: ./.github/actions/setup-api-client
+        with:
+          secrets: ${{ toJSON(secrets) }}
+          github_token: ${{ github.token }}
       - name: Create issue on repeated failures
         env:
           GH_TOKEN: ${{ github.token }}
@@ -1692,6 +1701,9 @@ jobs:
         uses: actions/checkout@v6
       - name: Export load-balancer tokens
         uses: ./.github/actions/setup-api-client
+        with:
+          secrets: ${{ toJSON(secrets) }}
+          github_token: ${{ github.token }}
       - name: Parse date range
         id: dates
         run: |
diff --git a/.github/workflows/reusable-codex-run.yml b/.github/workflows/reusable-codex-run.yml
index 3a9e8c18c..e2221e6b6 100644
--- a/.github/workflows/reusable-codex-run.yml
+++ b/.github/workflows/reusable-codex-run.yml
@@ -1018,16 +1018,16 @@ jobs:
 
           # Extract key fields for downstream use
           if [ -f "$ANALYSIS_FILE" ]; then
-            COMPLETED=$(python3 -c "import json; d=json.load(open('$ANALYSIS_FILE')); print(json.dumps(d.get('completed_tasks', [])))")
-            PROVIDER=$(python3 -c "import json; d=json.load(open('$ANALYSIS_FILE')); print(d.get('provider', 'unknown'))")
-            MODEL=$(python3 -c "import json; d=json.load(open('$ANALYSIS_FILE')); print(d.get('model', 'unknown'))")
-            CONFIDENCE=$(python3 -c "import json; d=json.load(open('$ANALYSIS_FILE')); print(d.get('confidence', 0))")
+            COMPLETED=$(python3 -c "import json; d=json.load(open('$ANALYSIS_FILE')); print(json.dumps(d.get('completed_tasks', [])), end='')")
+            PROVIDER=$(python3 -c "import json; d=json.load(open('$ANALYSIS_FILE')); print(d.get('provider', 'unknown'), end='')")
+            MODEL=$(python3 -c "import json; d=json.load(open('$ANALYSIS_FILE')); print(d.get('model', 'unknown'), end='')")
+            CONFIDENCE=$(python3 -c "import json; d=json.load(open('$ANALYSIS_FILE')); print(d.get('confidence', 0), end='')")
             # Quality metrics for keepalive integration
-            RAW_CONFIDENCE=$(python3 -c "import json; d=json.load(open('$ANALYSIS_FILE')); print(d.get('raw_confidence', d.get('confidence', 0)))")
-            EFFORT_SCORE=$(python3 -c "import json; d=json.load(open('$ANALYSIS_FILE')); print(d.get('effort_score', 0))")
-            DATA_QUALITY=$(python3 -c "import json; d=json.load(open('$ANALYSIS_FILE')); print(d.get('data_quality', 'unknown'))")
-            ANALYSIS_LENGTH=$(python3 -c "import json; d=json.load(open('$ANALYSIS_FILE')); print(d.get('analysis_text_length', 0))")
-            QUALITY_WARNINGS=$(python3 -c "import json; d=json.load(open('$ANALYSIS_FILE')); print(json.dumps(d.get('quality_warnings', [])))")
+            RAW_CONFIDENCE=$(python3 -c "import json; d=json.load(open('$ANALYSIS_FILE')); print(d.get('raw_confidence', d.get('confidence', 0)), end='')")
+            EFFORT_SCORE=$(python3 -c "import json; d=json.load(open('$ANALYSIS_FILE')); print(d.get('effort_score', 0), end='')")
+            DATA_QUALITY=$(python3 -c "import json; d=json.load(open('$ANALYSIS_FILE')); print(d.get('data_quality', 'unknown'), end='')")
+            ANALYSIS_LENGTH=$(python3 -c "import json; d=json.load(open('$ANALYSIS_FILE')); print(d.get('analysis_text_length', 0), end='')")
+            QUALITY_WARNINGS=$(python3 -c "import json; d=json.load(open('$ANALYSIS_FILE')); print(json.dumps(d.get('quality_warnings', [])), end='')")
             {
               echo "completed-tasks=$COMPLETED"
               echo "provider=$PROVIDER"
@@ -1185,7 +1185,7 @@ jobs:
 
           # Check if there are any staged changes remaining after reset
           if git diff --cached --quiet; then
-            echo "::warning::No non-artifact changes to commit after filtering. All changes were workflow artifacts."
+            echo "::notice::No non-artifact changes to commit after filtering. All changes were workflow artifacts."
             # Even if nothing to commit, check for unpushed commits that need pushing
             git fetch "${REMOTE_URL}" "${TARGET_BRANCH}" 2>/dev/null || true
             UNPUSHED=$(git rev-list FETCH_HEAD..HEAD --count 2>/dev/null || echo "0")
diff --git a/templates/consumer-repo/.github/scripts/error_classifier.js b/templates/consumer-repo/.github/scripts/error_classifier.js
index 18f89c2c3..85709c43f 100644
--- a/templates/consumer-repo/.github/scripts/error_classifier.js
+++ b/templates/consumer-repo/.github/scripts/error_classifier.js
@@ -189,8 +189,10 @@ function classifyByMessage(message) {
 function classifyError(error) {
   const message = normaliseMessage(error);
   const preview = message ? message.slice(0, 50) : 'unknown';
-  // eslint-disable-next-line no-console
-  console.log(`[error_classifier] Classifying error: ${preview}`);
+  if (process.env.RUNNER_DEBUG === '1') {
+    // eslint-disable-next-line no-console
+    console.log(`[error_classifier] Classifying error: ${preview}`);
+  }
   const status = getStatusCode(error);
 
   const statusCategory = status ? classifyByStatus(status, message) : null;
diff --git a/templates/consumer-repo/.github/scripts/keepalive_loop.js b/templates/consumer-repo/.github/scripts/keepalive_loop.js
index 26938a680..96cdf80e4 100644
--- a/templates/consumer-repo/.github/scripts/keepalive_loop.js
+++ b/templates/consumer-repo/.github/scripts/keepalive_loop.js
@@ -374,7 +374,7 @@ async function fetchRepoVariables({ github, context, core, names = [] }) {
     }
   } catch (error) {
     if (core) {
-      core.info(`Failed to fetch repository variables for timeout config: ${error.message}`);
+      core.debug(`Repository variables not accessible for timeout config (using defaults): ${error.message}`);
     }
   }
 
diff --git a/templates/consumer-repo/.github/workflows/agents-72-codex-belt-worker.yml b/templates/consumer-repo/.github/workflows/agents-72-codex-belt-worker.yml
index 1c4dcd0a3..87c148c82 100644
--- a/templates/consumer-repo/.github/workflows/agents-72-codex-belt-worker.yml
+++ b/templates/consumer-repo/.github/workflows/agents-72-codex-belt-worker.yml
@@ -591,6 +591,13 @@ jobs:
           fetch-depth: 1
           path: .belt-tools
 
+      - name: Re-install API client after branch checkout
+        if: ${{ steps.parallel.outputs.allowed == 'true' && (inputs.keepalive != true || steps.keepalive_gate.outputs.action != 'skip') }}
+        uses: ./.github/actions/setup-api-client
+        with:
+          secrets: ${{ toJSON(secrets) }}
+          github_token: ${{ github.token }}
+
       - name: Validate ledger base branch
         if: ${{ steps.parallel.outputs.allowed == 'true' && (inputs.keepalive != true || steps.keepalive_gate.outputs.action != 'skip') }}
         env:
@@ -764,7 +771,7 @@ jobs:
                   return super().increase_indent(flow, False)
 
           def iso_now() -> str:
-              return dt.datetime.utcnow().replace(microsecond=0).isoformat() + 'Z'
+              return dt.datetime.now(dt.timezone.utc).replace(microsecond=0).isoformat().replace('+00:00', 'Z')
 
           issue = os.environ["ISSUE"].strip()
           branch = os.environ["BRANCH"].strip()
@@ -1272,7 +1279,7 @@ jobs:
                   return super().increase_indent(flow, False)
 
           def iso_now() -> str:
-              return dt.datetime.utcnow().replace(microsecond=0).isoformat() + 'Z'
+              return dt.datetime.now(dt.timezone.utc).replace(microsecond=0).isoformat().replace('+00:00', 'Z')
 
           issue = os.environ.get('ISSUE', '').strip()
           branch = os.environ.get('BRANCH', '').strip()

From 926a8dcf8a1af13304f5e8cf4ab1f20a5802401a Mon Sep 17 00:00:00 2001
From: stranske <stranske@gmail.com>
Date: Thu, 12 Feb 2026 01:26:13 +0000
Subject: [PATCH 02/11] fix: address review comments on belt worker re-install
 step

1. Use .belt-tools action path instead of ./ for setup-api-client
   after branch checkout, so the action runs from trusted Workflows
   code rather than the untrusted issue branch (security fix).

2. Pass GH_BELT_TOKEN || github.token as github_token input to
   preserve the belt token selection instead of overriding
   GITHUB_TOKEN/GH_TOKEN with the default workflow token.
---
 .github/workflows/agents-72-codex-belt-worker.yml             | 4 ++--
 .../.github/workflows/agents-72-codex-belt-worker.yml         | 4 ++--
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/.github/workflows/agents-72-codex-belt-worker.yml b/.github/workflows/agents-72-codex-belt-worker.yml
index 95530873d..ec87d6d2f 100644
--- a/.github/workflows/agents-72-codex-belt-worker.yml
+++ b/.github/workflows/agents-72-codex-belt-worker.yml
@@ -593,10 +593,10 @@ jobs:
 
       - name: Re-install API client after branch checkout
         if: ${{ steps.parallel.outputs.allowed == 'true' && (inputs.keepalive != true || steps.keepalive_gate.outputs.action != 'skip') }}
-        uses: ./.github/actions/setup-api-client
+        uses: ./.belt-tools/.github/actions/setup-api-client
         with:
           secrets: ${{ toJSON(secrets) }}
-          github_token: ${{ github.token }}
+          github_token: ${{ env.GH_BELT_TOKEN || github.token }}
 
       - name: Validate ledger base branch
         if: ${{ steps.parallel.outputs.allowed == 'true' && (inputs.keepalive != true || steps.keepalive_gate.outputs.action != 'skip') }}
diff --git a/templates/consumer-repo/.github/workflows/agents-72-codex-belt-worker.yml b/templates/consumer-repo/.github/workflows/agents-72-codex-belt-worker.yml
index 87c148c82..14e6f2bfb 100644
--- a/templates/consumer-repo/.github/workflows/agents-72-codex-belt-worker.yml
+++ b/templates/consumer-repo/.github/workflows/agents-72-codex-belt-worker.yml
@@ -593,10 +593,10 @@ jobs:
 
       - name: Re-install API client after branch checkout
         if: ${{ steps.parallel.outputs.allowed == 'true' && (inputs.keepalive != true || steps.keepalive_gate.outputs.action != 'skip') }}
-        uses: ./.github/actions/setup-api-client
+        uses: ./.belt-tools/.github/actions/setup-api-client
         with:
           secrets: ${{ toJSON(secrets) }}
-          github_token: ${{ github.token }}
+          github_token: ${{ env.GH_BELT_TOKEN || github.token }}
 
       - name: Validate ledger base branch
         if: ${{ steps.parallel.outputs.allowed == 'true' && (inputs.keepalive != true || steps.keepalive_gate.outputs.action != 'skip') }}

From 24d84ec89f9038bfa511f4e2af0ea1f3d105e3ec Mon Sep 17 00:00:00 2001
From: stranske <stranske@gmail.com>
Date: Thu, 12 Feb 2026 02:41:46 +0000
Subject: [PATCH 03/11] fix: capability_check false-positive on 'secrets' +
 lower verdict threshold

Two independent fixes for broken automation flows:

1. capability_check.py: The bare \bsecrets?\b regex matched negative
   mentions like 'no secrets' in issue constraint text, causing
   _requires_admin_access() to return true and the fallback classifier
   to BLOCK tasks that merely *describe* a no-secrets constraint.
   Replace with specific verb+secrets patterns (manage/configure/set/
   create/update/delete/add/modify/rotate secrets).
   Root cause of PAEM #1403 false-positive BLOCKED.

2. verdict_policy.py: CONCERNS_NEEDS_HUMAN_THRESHOLD lowered from 0.85
   to 0.50.  The old threshold meant any split verdict (PASS + CONCERNS)
   with <85% confidence on the concerns side triggered needs_human,
   blocking automatic follow-up issue creation.  A 72% confidence
   concerns verdict (TMP #4894) is well above chance and should produce
   a follow-up rather than require manual triage.

Both template and main copies updated; new regression tests added.
---
 scripts/langchain/capability_check.py         |  3 +-
 scripts/langchain/verdict_policy.py           |  2 +-
 .../scripts/langchain/capability_check.py     |  3 +-
 tests/scripts/test_capability_check.py        | 15 ++++++++
 tests/test_verdict_policy.py                  | 34 +++++++++++++++++++
 5 files changed, 54 insertions(+), 3 deletions(-)

diff --git a/scripts/langchain/capability_check.py b/scripts/langchain/capability_check.py
index eb904c753..0ff398491 100755
--- a/scripts/langchain/capability_check.py
+++ b/scripts/langchain/capability_check.py
@@ -161,7 +161,8 @@ def _is_multi_action_task(task: str) -> bool:
 def _requires_admin_access(task: str) -> bool:
     patterns = [
         r"\bgithub\s+secrets?\b",
-        r"\bsecrets?\b",
+        r"\b(?:manage|configure|set|create|update|delete|add|modify|rotate)\s+secrets?\b",
+        r"\bsecrets?\s+(?:management|configuration|rotation)\b",
         r"\brepository\s+settings\b",
         r"\brepo\s+settings\b",
         r"\bbranch\s+protection\b",
diff --git a/scripts/langchain/verdict_policy.py b/scripts/langchain/verdict_policy.py
index 8fdcd9925..809bdcadb 100644
--- a/scripts/langchain/verdict_policy.py
+++ b/scripts/langchain/verdict_policy.py
@@ -16,7 +16,7 @@
     "fail": 3,
 }
 
-CONCERNS_NEEDS_HUMAN_THRESHOLD = 0.85
+CONCERNS_NEEDS_HUMAN_THRESHOLD = 0.50
 
 
 @dataclass(frozen=True)
diff --git a/templates/consumer-repo/scripts/langchain/capability_check.py b/templates/consumer-repo/scripts/langchain/capability_check.py
index 0632d893d..b53886821 100755
--- a/templates/consumer-repo/scripts/langchain/capability_check.py
+++ b/templates/consumer-repo/scripts/langchain/capability_check.py
@@ -160,7 +160,8 @@ def _is_multi_action_task(task: str) -> bool:
 def _requires_admin_access(task: str) -> bool:
     patterns = [
         r"\bgithub\s+secrets?\b",
-        r"\bsecrets?\b",
+        r"\b(?:manage|configure|set|create|update|delete|add|modify|rotate)\s+secrets?\b",
+        r"\bsecrets?\s+(?:management|configuration|rotation)\b",
         r"\brepository\s+settings\b",
         r"\brepo\s+settings\b",
         r"\bbranch\s+protection\b",
diff --git a/tests/scripts/test_capability_check.py b/tests/scripts/test_capability_check.py
index 28f5233c6..e35788188 100644
--- a/tests/scripts/test_capability_check.py
+++ b/tests/scripts/test_capability_check.py
@@ -419,6 +419,21 @@ def test_fallback_flags_admin_requirement(self) -> None:
             assert result.blocked_tasks[0]["task"] == "Update GitHub secrets"
             assert "admin" in result.blocked_tasks[0]["reason"].lower()
 
+    def test_fallback_does_not_flag_negated_secrets_mention(self) -> None:
+        """Regression: 'no secrets' in constraint text must not trigger admin block."""
+        task = "safety rules (no secrets, no workflow edits, no file writes)"
+        with mock.patch("scripts.langchain.capability_check._get_llm_client", return_value=None):
+            result = classify_capabilities([task], "")
+            assert result.recommendation != "BLOCKED"
+            assert all(item["task"] != task for item in result.blocked_tasks)
+
+    def test_fallback_flags_manage_secrets(self) -> None:
+        """Specific secrets-management verbs should still be blocked."""
+        with mock.patch("scripts.langchain.capability_check._get_llm_client", return_value=None):
+            result = classify_capabilities(["manage secrets for deployment"], "")
+            assert result.recommendation == "BLOCKED"
+            assert "admin" in result.blocked_tasks[0]["reason"].lower()
+
     def test_fallback_suggests_decomposition(self) -> None:
         with mock.patch("scripts.langchain.capability_check._get_llm_client", return_value=None):
             result = classify_capabilities(["Refactor auth + add tests + update docs"], "")
diff --git a/tests/test_verdict_policy.py b/tests/test_verdict_policy.py
index 9b5075406..7e134a5d5 100755
--- a/tests/test_verdict_policy.py
+++ b/tests/test_verdict_policy.py
@@ -61,3 +61,37 @@ def test_needs_human_threshold_boundary():
     result = evaluate_verdict_policy(verdicts, policy="worst")
 
     assert result.needs_human is False
+
+
+def test_needs_human_true_below_threshold():
+    """Concerns below the threshold should trigger needs_human."""
+    verdicts = [
+        ProviderVerdict("openai", "gpt-5.2", "PASS", 0.92),
+        ProviderVerdict("anthropic", "claude-sonnet-4-5", "CONCERNS", 0.40),
+    ]
+
+    result = evaluate_verdict_policy(verdicts, policy="worst")
+
+    assert result.needs_human is True
+    assert result.split_verdict is True
+    assert "low-confidence" in result.needs_human_reason
+
+
+def test_moderate_confidence_concerns_do_not_block():
+    """Regression: 72% concerns in a split verdict should not trigger needs_human.
+
+    Previously CONCERNS_NEEDS_HUMAN_THRESHOLD was 0.85, which caused any
+    split verdict with <85% concerns to be flagged.  The lowered threshold
+    (0.50) allows moderate-confidence concerns to proceed with automatic
+    follow-up creation.
+    """
+    verdicts = [
+        ProviderVerdict("openai", "gpt-5.2", "CONCERNS", 72),
+        ProviderVerdict("anthropic", "claude-sonnet-4-5", "PASS", 85),
+    ]
+
+    result = evaluate_verdict_policy(verdicts, policy="worst")
+
+    assert result.split_verdict is True
+    assert result.needs_human is False
+    assert result.verdict == "CONCERNS"

From 00365f43e3f010729548560f4b6b196afa48b4ef Mon Sep 17 00:00:00 2001
From: stranske <stranske@gmail.com>
Date: Thu, 12 Feb 2026 12:48:51 +0000
Subject: [PATCH 04/11] fix: prevent Codex bootstrap from overwriting vendored
 node_modules
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Three-layer fix for the systemic issue where setup-api-client's npm install
overwrites vendored minimatch package.json, and git add -A captures the
modification into bootstrap/autofix commits.

Layer 1 (source fix): setup-api-client/action.yml
  - Snapshot vendored package.json files before npm install
  - Restore them after npm install completes
  - Applied to both .github/actions/ and templates/consumer-repo/

Layer 2 (targeted staging): reusable-agents-issue-bridge.yml
  - Replace 'git add -A' with targeted 'git add agents/${AGENT}-${ISSUE}.md'
  - Only the bootstrap file gets staged, not npm side-effects

Layer 3 (safety net): reusable-18-autofix.yml
  - Add 'git reset HEAD -- .github/scripts/node_modules ...' after git add -A
  - Matches existing pattern in reusable-codex-run.yml line 1184
  - Applied to both push-commit and patch-commit paths

Also fixes test assertions that referenced the old CONCERNS_NEEDS_HUMAN_THRESHOLD
(was 0.85, now 0.50) — confidence values in tests updated accordingly.

Fixes: Copilot review finding on PAEM PR #1417 (minimatch vendoring cycle)
---
 .github/actions/setup-api-client/action.yml   | 30 +++++++++++++++++++
 .github/workflows/reusable-18-autofix.yml     |  4 +++
 .../reusable-agents-issue-bridge.yml          |  4 ++-
 .../actions/setup-api-client/action.yml       | 30 +++++++++++++++++++
 tests/test_followup_issue_generator.py        |  2 +-
 tests/test_verdict_extract.py                 |  2 +-
 tests/test_verdict_policy_integration.py      |  2 +-
 7 files changed, 70 insertions(+), 4 deletions(-)

diff --git a/.github/actions/setup-api-client/action.yml b/.github/actions/setup-api-client/action.yml
index 3343ebfb5..109cc33a7 100644
--- a/.github/actions/setup-api-client/action.yml
+++ b/.github/actions/setup-api-client/action.yml
@@ -102,6 +102,22 @@ runs:
         if [ -d "node_modules/@octokit/rest" ]; then
           echo "✅ @octokit/rest already installed"
         else
+          # Snapshot vendored package metadata before npm install.
+          # npm may overwrite transitive deps (e.g. minimatch) that are
+          # committed as vendored packages with intentional version pins.
+          VENDORED_SNAPSHOT=""
+          if [ -f "node_modules/minimatch/package.json" ]; then
+            VENDORED_SNAPSHOT=$(mktemp -d)
+            for pkg_dir in node_modules/*/; do
+              if [ -f "${pkg_dir}package.json" ]; then
+                pkg_name=$(basename "$pkg_dir")
+                mkdir -p "${VENDORED_SNAPSHOT}/${pkg_name}"
+                cp "${pkg_dir}package.json" "${VENDORED_SNAPSHOT}/${pkg_name}/package.json"
+              fi
+            done
+            echo "📸 Snapshotted vendored package metadata"
+          fi
+
           # Install with pinned versions for consistency
           # Capture stderr for debugging if the command fails
           npm_output=$(mktemp)
@@ -122,6 +138,20 @@ runs:
               @octokit/plugin-paginate-rest@9.1.5 \
               @octokit/auth-app@6.0.3
           fi
+
+          # Restore vendored package metadata that npm may have overwritten
+          if [ -n "${VENDORED_SNAPSHOT:-}" ] && [ -d "${VENDORED_SNAPSHOT}" ]; then
+            for pkg_backup in "${VENDORED_SNAPSHOT}"/*/; do
+              pkg_name=$(basename "$pkg_backup")
+              if [ -f "node_modules/${pkg_name}/package.json" ] && \
+                 [ -f "${pkg_backup}package.json" ]; then
+                cp "${pkg_backup}package.json" "node_modules/${pkg_name}/package.json"
+              fi
+            done
+            rm -rf "${VENDORED_SNAPSHOT}"
+            echo "📸 Restored vendored package metadata"
+          fi
+
           echo "✅ @octokit dependencies installed"
         fi
 
diff --git a/.github/workflows/reusable-18-autofix.yml b/.github/workflows/reusable-18-autofix.yml
index d4d64636a..f24a6f975 100644
--- a/.github/workflows/reusable-18-autofix.yml
+++ b/.github/workflows/reusable-18-autofix.yml
@@ -780,6 +780,8 @@ jobs:
           git config user.name "github-actions[bot]"
           git config user.email "github-actions[bot]@users.noreply.github.com"
           git add -A
+          # Unstage vendored node_modules that may have been modified by npm install
+          git reset HEAD -- .github/scripts/node_modules node_modules .workflows-lib/.github/scripts/node_modules 2>/dev/null || true
           git commit -m "${AUTOFIX_COMMIT_PREFIX} formatting/lint"
           echo "AUTOFIX_COMMIT_SHA=$(git rev-parse HEAD)" >> "$GITHUB_ENV"
 
@@ -851,6 +853,8 @@ jobs:
           git config user.name "github-actions[bot]"
           git config user.email "github-actions[bot]@users.noreply.github.com"
           git add -A
+          # Unstage vendored node_modules that may have been modified by npm install
+          git reset HEAD -- .github/scripts/node_modules node_modules .workflows-lib/.github/scripts/node_modules 2>/dev/null || true
           git commit -m "${AUTOFIX_COMMIT_PREFIX} formatting/lint (patch)" || true
           git format-patch -1 --stdout > autofix.patch
 
diff --git a/.github/workflows/reusable-agents-issue-bridge.yml b/.github/workflows/reusable-agents-issue-bridge.yml
index d193a16ce..16dea6053 100644
--- a/.github/workflows/reusable-agents-issue-bridge.yml
+++ b/.github/workflows/reusable-agents-issue-bridge.yml
@@ -580,7 +580,9 @@ jobs:
           git checkout -B "$HEAD_BRANCH" "origin/${BASE_BRANCH}"
           mkdir -p agents
           printf "<!-- bootstrap for %s on issue #%s -->\n" "$AGENT" "$ISSUE_NUM" > "agents/${AGENT}-${ISSUE_NUM}.md"
-          git add -A || true
+          # Stage only the intended bootstrap file — 'git add -A' would capture
+          # vendored node_modules changes made by setup-api-client npm install.
+          git add "agents/${AGENT}-${ISSUE_NUM}.md" || true
           if ! git diff --cached --quiet; then
             git commit -m "chore(${AGENT}): bootstrap PR for issue #${ISSUE_NUM}"
           else
diff --git a/templates/consumer-repo/.github/actions/setup-api-client/action.yml b/templates/consumer-repo/.github/actions/setup-api-client/action.yml
index 3343ebfb5..109cc33a7 100644
--- a/templates/consumer-repo/.github/actions/setup-api-client/action.yml
+++ b/templates/consumer-repo/.github/actions/setup-api-client/action.yml
@@ -102,6 +102,22 @@ runs:
         if [ -d "node_modules/@octokit/rest" ]; then
           echo "✅ @octokit/rest already installed"
         else
+          # Snapshot vendored package metadata before npm install.
+          # npm may overwrite transitive deps (e.g. minimatch) that are
+          # committed as vendored packages with intentional version pins.
+          VENDORED_SNAPSHOT=""
+          if [ -f "node_modules/minimatch/package.json" ]; then
+            VENDORED_SNAPSHOT=$(mktemp -d)
+            for pkg_dir in node_modules/*/; do
+              if [ -f "${pkg_dir}package.json" ]; then
+                pkg_name=$(basename "$pkg_dir")
+                mkdir -p "${VENDORED_SNAPSHOT}/${pkg_name}"
+                cp "${pkg_dir}package.json" "${VENDORED_SNAPSHOT}/${pkg_name}/package.json"
+              fi
+            done
+            echo "📸 Snapshotted vendored package metadata"
+          fi
+
           # Install with pinned versions for consistency
           # Capture stderr for debugging if the command fails
           npm_output=$(mktemp)
@@ -122,6 +138,20 @@ runs:
               @octokit/plugin-paginate-rest@9.1.5 \
               @octokit/auth-app@6.0.3
           fi
+
+          # Restore vendored package metadata that npm may have overwritten
+          if [ -n "${VENDORED_SNAPSHOT:-}" ] && [ -d "${VENDORED_SNAPSHOT}" ]; then
+            for pkg_backup in "${VENDORED_SNAPSHOT}"/*/; do
+              pkg_name=$(basename "$pkg_backup")
+              if [ -f "node_modules/${pkg_name}/package.json" ] && \
+                 [ -f "${pkg_backup}package.json" ]; then
+                cp "${pkg_backup}package.json" "node_modules/${pkg_name}/package.json"
+              fi
+            done
+            rm -rf "${VENDORED_SNAPSHOT}"
+            echo "📸 Restored vendored package metadata"
+          fi
+
           echo "✅ @octokit dependencies installed"
         fi
 
diff --git a/tests/test_followup_issue_generator.py b/tests/test_followup_issue_generator.py
index 572099608..aeb9d1d38 100755
--- a/tests/test_followup_issue_generator.py
+++ b/tests/test_followup_issue_generator.py
@@ -533,7 +533,7 @@ def test_split_low_confidence_requires_needs_human(self):
         verification_data = VerificationData(
             provider_verdicts={
                 "openai": {"verdict": "PASS", "confidence": 90},
-                "anthropic": {"verdict": "CONCERNS", "confidence": 70},
+                "anthropic": {"verdict": "CONCERNS", "confidence": 40},
             },
             concerns=["Missing test coverage"],
         )
diff --git a/tests/test_verdict_extract.py b/tests/test_verdict_extract.py
index aa3a207cc..938df898a 100644
--- a/tests/test_verdict_extract.py
+++ b/tests/test_verdict_extract.py
@@ -25,7 +25,7 @@ def _parse_github_output(raw: str) -> dict[str, str]:
 def test_verdict_extract_emits_structured_github_outputs(tmp_path):
     summary = _build_summary(
         "| openai | gpt-5.2 | PASS | 0.92 | Looks good. |",
-        "| anthropic | claude-sonnet-4-5 | CONCERNS | 0.84 | Missing edge case. |",
+        "| anthropic | claude-sonnet-4-5 | CONCERNS | 0.40 | Missing edge case. |",
     )
     result = verdict_extract.build_verdict_result(summary, policy="worst")
     output_path = tmp_path / "github_output.txt"
diff --git a/tests/test_verdict_policy_integration.py b/tests/test_verdict_policy_integration.py
index 67afa4ee5..1cca00d00 100755
--- a/tests/test_verdict_policy_integration.py
+++ b/tests/test_verdict_policy_integration.py
@@ -40,7 +40,7 @@ def test_split_verdict_confidence_boundary_needs_human_false():
 def test_split_verdict_low_confidence_needs_human_true():
     summary = _build_summary(
         "| openai | gpt-5.2 | PASS | 0.92 | Looks good. |",
-        "| anthropic | claude-sonnet-4-5 | CONCERNS | 0.84 | Missing edge case. |",
+        "| anthropic | claude-sonnet-4-5 | CONCERNS | 0.40 | Missing edge case. |",
     )
 
     workflow_result = _workflow_result(summary)

From 089bc23fa925bb7dc47e9655bacf4d7335f6a8a5 Mon Sep 17 00:00:00 2001
From: stranske <stranske@gmail.com>
Date: Thu, 12 Feb 2026 12:58:14 +0000
Subject: [PATCH 05/11] fix: flip needs_human to trigger on high-confidence
 CONCERNS, not low
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The needs_human gate was backwards: it fired when the CONCERNS provider
had LOW confidence (LLM unsure there's a problem) instead of HIGH
confidence (LLM confident there's a real problem).

Confidence reflects the LLM's certainty in its own evaluation, not a
measure of code quality. Low-confidence CONCERNS is a weak signal that
shouldn't block follow-up automation. High-confidence CONCERNS is the
stronger signal warranting human review.

Changed: confidence_value < threshold  →  confidence_value >= threshold
Threshold set to 0.85 (high bar — a human is already in the loop and
depth-of-rounds provides an independent guard against runaway automation).
---
 scripts/langchain/verdict_policy.py      |  8 ++++----
 tests/test_followup_issue_generator.py   |  6 +++---
 tests/test_verdict_extract.py            |  2 +-
 tests/test_verdict_policy.py             | 21 +++++++++++----------
 tests/test_verdict_policy_integration.py | 14 ++++++++------
 5 files changed, 27 insertions(+), 24 deletions(-)

diff --git a/scripts/langchain/verdict_policy.py b/scripts/langchain/verdict_policy.py
index 809bdcadb..86d422350 100644
--- a/scripts/langchain/verdict_policy.py
+++ b/scripts/langchain/verdict_policy.py
@@ -16,7 +16,7 @@
     "fail": 3,
 }
 
-CONCERNS_NEEDS_HUMAN_THRESHOLD = 0.50
+CONCERNS_NEEDS_HUMAN_THRESHOLD = 0.85
 
 
 @dataclass(frozen=True)
@@ -193,11 +193,11 @@ def evaluate_verdict_policy(
     needs_human_reason = ""
     if split_verdict:
         confidence_value = concerns_confidence or 0.0
-        if confidence_value < CONCERNS_NEEDS_HUMAN_THRESHOLD:
+        if confidence_value >= CONCERNS_NEEDS_HUMAN_THRESHOLD:
             needs_human = True
             needs_human_reason = (
-                "Provider verdicts split with low-confidence concerns; "
-                f"dissenting confidence {confidence_value:.2f} < "
+                "Provider verdicts split with high-confidence concerns; "
+                f"dissenting confidence {confidence_value:.2f} >= "
                 f"{CONCERNS_NEEDS_HUMAN_THRESHOLD:.2f}. "
                 "Requires human review before starting another automated follow-up."
             )
diff --git a/tests/test_followup_issue_generator.py b/tests/test_followup_issue_generator.py
index aeb9d1d38..c3f613be6 100755
--- a/tests/test_followup_issue_generator.py
+++ b/tests/test_followup_issue_generator.py
@@ -528,12 +528,12 @@ def test_advisory_concerns_are_notes(self):
         assert "- [ ] Address: Could add a clarifying comment" not in followup.body
         assert "## Notes" in followup.body
 
-    def test_split_low_confidence_requires_needs_human(self):
-        """Low-confidence split verdicts should trigger needs-human labeling."""
+    def test_split_high_confidence_requires_needs_human(self):
+        """High-confidence CONCERNS in a split verdict should trigger needs-human labeling."""
         verification_data = VerificationData(
             provider_verdicts={
                 "openai": {"verdict": "PASS", "confidence": 90},
-                "anthropic": {"verdict": "CONCERNS", "confidence": 40},
+                "anthropic": {"verdict": "CONCERNS", "confidence": 92},
             },
             concerns=["Missing test coverage"],
         )
diff --git a/tests/test_verdict_extract.py b/tests/test_verdict_extract.py
index 938df898a..38d43456a 100644
--- a/tests/test_verdict_extract.py
+++ b/tests/test_verdict_extract.py
@@ -25,7 +25,7 @@ def _parse_github_output(raw: str) -> dict[str, str]:
 def test_verdict_extract_emits_structured_github_outputs(tmp_path):
     summary = _build_summary(
         "| openai | gpt-5.2 | PASS | 0.92 | Looks good. |",
-        "| anthropic | claude-sonnet-4-5 | CONCERNS | 0.40 | Missing edge case. |",
+        "| anthropic | claude-sonnet-4-5 | CONCERNS | 0.90 | Missing edge case. |",
     )
     result = verdict_extract.build_verdict_result(summary, policy="worst")
     output_path = tmp_path / "github_output.txt"
diff --git a/tests/test_verdict_policy.py b/tests/test_verdict_policy.py
index 7e134a5d5..bfe631e05 100755
--- a/tests/test_verdict_policy.py
+++ b/tests/test_verdict_policy.py
@@ -51,6 +51,7 @@ def test_select_verdict_majority_policy():
 
 
 def test_needs_human_threshold_boundary():
+    """At exactly the threshold, needs_human should fire (>= comparison)."""
     verdicts = [
         ProviderVerdict("openai", "gpt-5.2", "PASS", 0.92),
         ProviderVerdict(
@@ -60,30 +61,30 @@ def test_needs_human_threshold_boundary():
 
     result = evaluate_verdict_policy(verdicts, policy="worst")
 
-    assert result.needs_human is False
+    assert result.needs_human is True
 
 
-def test_needs_human_true_below_threshold():
-    """Concerns below the threshold should trigger needs_human."""
+def test_needs_human_true_above_threshold():
+    """Concerns above the threshold should trigger needs_human."""
     verdicts = [
         ProviderVerdict("openai", "gpt-5.2", "PASS", 0.92),
-        ProviderVerdict("anthropic", "claude-sonnet-4-5", "CONCERNS", 0.40),
+        ProviderVerdict("anthropic", "claude-sonnet-4-5", "CONCERNS", 0.90),
     ]
 
     result = evaluate_verdict_policy(verdicts, policy="worst")
 
     assert result.needs_human is True
     assert result.split_verdict is True
-    assert "low-confidence" in result.needs_human_reason
+    assert "high-confidence" in result.needs_human_reason
 
 
 def test_moderate_confidence_concerns_do_not_block():
-    """Regression: 72% concerns in a split verdict should not trigger needs_human.
+    """Moderate-confidence concerns in a split verdict should not trigger needs_human.
 
-    Previously CONCERNS_NEEDS_HUMAN_THRESHOLD was 0.85, which caused any
-    split verdict with <85% concerns to be flagged.  The lowered threshold
-    (0.50) allows moderate-confidence concerns to proceed with automatic
-    follow-up creation.
+    needs_human only fires when the CONCERNS provider is highly confident
+    (>= 0.85), indicating the LLM is quite sure there are real problems.
+    Moderate confidence means the LLM is uncertain — that's a weaker signal
+    and shouldn't block follow-up automation.
     """
     verdicts = [
         ProviderVerdict("openai", "gpt-5.2", "CONCERNS", 72),
diff --git a/tests/test_verdict_policy_integration.py b/tests/test_verdict_policy_integration.py
index 1cca00d00..40562e48f 100755
--- a/tests/test_verdict_policy_integration.py
+++ b/tests/test_verdict_policy_integration.py
@@ -23,7 +23,8 @@ def _build_summary(*rows: str) -> str:
     return f"## Provider Summary\n\n{header}{body}\n"
 
 
-def test_split_verdict_confidence_boundary_needs_human_false():
+def test_split_verdict_confidence_boundary_needs_human_true():
+    """At exactly the threshold (0.85), needs_human should be True."""
     summary = _build_summary(
         "| openai | gpt-5.2 | PASS | 0.92 | Looks good. |",
         "| anthropic | claude-sonnet-4-5 | CONCERNS | 0.85 | Missing edge case. |",
@@ -33,11 +34,12 @@ def test_split_verdict_confidence_boundary_needs_human_false():
     followup_result = _followup_result(summary)
 
     assert workflow_result.verdict == followup_result.verdict == "CONCERNS"
-    assert workflow_result.needs_human is False
-    assert followup_result.needs_human is False
+    assert workflow_result.needs_human is True
+    assert followup_result.needs_human is True
 
 
-def test_split_verdict_low_confidence_needs_human_true():
+def test_split_verdict_low_confidence_needs_human_false():
+    """Below threshold, low-confidence concerns should NOT trigger needs_human."""
     summary = _build_summary(
         "| openai | gpt-5.2 | PASS | 0.92 | Looks good. |",
         "| anthropic | claude-sonnet-4-5 | CONCERNS | 0.40 | Missing edge case. |",
@@ -47,8 +49,8 @@ def test_split_verdict_low_confidence_needs_human_true():
     followup_result = _followup_result(summary)
 
     assert workflow_result.verdict == followup_result.verdict == "CONCERNS"
-    assert workflow_result.needs_human is True
-    assert followup_result.needs_human is True
+    assert workflow_result.needs_human is False
+    assert followup_result.needs_human is False
 
 
 def test_split_verdict_row_order_invariance():

From 371370e04b2f027f3b451ed6ba589e9d82f07ee4 Mon Sep 17 00:00:00 2001
From: stranske <stranske@gmail.com>
Date: Thu, 12 Feb 2026 15:23:28 +0000
Subject: [PATCH 06/11] =?UTF-8?q?fix:=20harden=20Codex=20pipeline=20?=
 =?UTF-8?q?=E2=80=94=20corrupt=20ledger=20resilience,=20autofix=20limits,?=
 =?UTF-8?q?=20task-focused=20prompts,=20PR=20meta=20debounce?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- ledger_migrate_base.py: skip corrupt YAML files instead of blocking all
  belt worker runs (root cause of issue #1418 stall)
- agents-autofix-loop: reduce max_attempts 3→2 (standard) and 2→1
  (escalated) to cut autofix churn observed in PR #4906
- agents-72-codex-belt-worker: emit task_title output and include
  task-focused directive in activation comment for higher first-commit
  success rate
- agents-pr-meta: add PR-number concurrency grouping with
  cancel-in-progress for pull_request events to debounce redundant runs
- All template counterparts updated in sync
- 2 new tests for corrupt ledger handling
---
 .../workflows/agents-72-codex-belt-worker.yml | 19 ++++++-
 .github/workflows/agents-autofix-loop.yml     |  4 +-
 .github/workflows/agents-pr-meta-v4.yml       |  4 +-
 scripts/ledger_migrate_base.py                | 14 ++++-
 .../workflows/agents-72-codex-belt-worker.yml | 19 ++++++-
 .../.github/workflows/agents-autofix-loop.yml |  4 +-
 .../.github/workflows/agents-pr-meta.yml      |  4 +-
 tests/scripts/test_ledger_migrate_base.py     | 51 ++++++++++++++++++-
 8 files changed, 108 insertions(+), 11 deletions(-)

diff --git a/.github/workflows/agents-72-codex-belt-worker.yml b/.github/workflows/agents-72-codex-belt-worker.yml
index ec87d6d2f..29a4dbefc 100644
--- a/.github/workflows/agents-72-codex-belt-worker.yml
+++ b/.github/workflows/agents-72-codex-belt-worker.yml
@@ -905,6 +905,7 @@ jobs:
           if gh_output:
               with open(gh_output, 'a', encoding='utf-8') as handle:
                   handle.write(f"task_id={start_info['task']['id'] if start_info['task'] else ''}\n")
+                  handle.write(f"task_title={start_info['task']['title'] if start_info['task'] else ''}\n")
                   handle.write(f"task_status={start_info['task']['current_status'] if start_info['task'] else ''}\n")
                   handle.write(f"ledger_changed={'true' if changed else 'false'}\n")
                   handle.write(f"ledger_created={'true' if start_info['created'] else 'false'}\n")
@@ -1187,12 +1188,28 @@ jobs:
             const prNumber = Number('${{ steps.pr.outputs.number }}');
             const branch = ('${{ steps.ctx.outputs.branch }}' || '').trim() || '(unknown branch)';
             const dryRun = '${{ steps.mode.outputs.dry_run }}' === 'true';
+            const taskId = ('${{ steps.ledger_start.outputs.task_id }}' || '').trim();
+            const taskTitle = ('${{ steps.ledger_start.outputs.task_title }}' || '').trim();
             const { owner, repo } = context.repo;
             const marker = '<!-- codex-activation-marker -->';
             const summary = dryRun
               ? `Codex Worker activated for branch \`${branch}\` (dry run preview).`
               : `Codex Worker activated for branch \`${branch}\`.`;
-            const body = `${marker}\n${summary}\n\n@codex start\n\nAutomated belt worker prepared this PR. Please continue implementing the requested changes.`;
+            // Direct Codex to focus on the single next ledger task for higher
+            // first-commit success probability.  Full issue context is in the
+            // PR body; this comment narrows the immediate scope.
+            let taskDirective = '';
+            if (taskId && taskTitle) {
+              taskDirective = [
+                '',
+                `**Focus on this task first:** \`${taskId}\` — ${taskTitle}`,
+                '',
+                'Implement **only** this task in your first commit.',
+                'Ensure the code compiles and existing tests pass before moving on.',
+                'The keepalive loop will assign subsequent tasks after this one is complete.',
+              ].join('\n');
+            }
+            const body = `${marker}\n${summary}\n\n@codex start${taskDirective}`;
 
             try {
               const comments = await paginateWithRetry(
diff --git a/.github/workflows/agents-autofix-loop.yml b/.github/workflows/agents-autofix-loop.yml
index f88518eb5..a989768c3 100644
--- a/.github/workflows/agents-autofix-loop.yml
+++ b/.github/workflows/agents-autofix-loop.yml
@@ -162,7 +162,7 @@ jobs:
               appendix: '',
               stop_reason: '',
               attempts: '0',
-              max_attempts: '3',
+              max_attempts: '2',
               trigger_reason: 'unknown',
               trigger_job: '',
               trigger_step: '',
@@ -287,7 +287,7 @@ jobs:
             // Reduce attempts for auto-escalated PRs (they weren't agent-initiated)
             const isEscalated = labels.includes('autofix:escalated');
             const maxAttempts = isEscalated
-              ? Math.min(2, Number(outputs.max_attempts))
+              ? 1
               : Number(outputs.max_attempts);
             const previousRuns = await paginateWithRetry(
               github,
diff --git a/.github/workflows/agents-pr-meta-v4.yml b/.github/workflows/agents-pr-meta-v4.yml
index 117b658f2..a7b4a7390 100644
--- a/.github/workflows/agents-pr-meta-v4.yml
+++ b/.github/workflows/agents-pr-meta-v4.yml
@@ -37,9 +37,11 @@ concurrency:
       && github.event.comment
       && github.event.comment.id
       && format('agents-pr-meta-comment-{0}', github.event.comment.id)
+      || github.event_name == 'pull_request'
+      && format('agents-pr-meta-pr-{0}', github.event.pull_request.number)
       || format('agents-pr-meta-run-{0}', github.run_id)
     }}
-  cancel-in-progress: false
+  cancel-in-progress: ${{ github.event_name == 'pull_request' }}
 
 jobs:
   comment_event_context:
diff --git a/scripts/ledger_migrate_base.py b/scripts/ledger_migrate_base.py
index 2a551ecac..a7fcadc8d 100644
--- a/scripts/ledger_migrate_base.py
+++ b/scripts/ledger_migrate_base.py
@@ -174,14 +174,26 @@ def main(argv: Iterable[str] | None = None) -> int:
 
     mismatches: list[LedgerResult] = []
     updated: list[LedgerResult] = []
+    skipped: list[tuple[Path, str]] = []
     for ledger_path in ledgers:
-        result = migrate_ledger(ledger_path, default_branch, check=args.check)
+        try:
+            result = migrate_ledger(ledger_path, default_branch, check=args.check)
+        except (MigrationError, yaml.YAMLError) as exc:
+            # One corrupt ledger must not block processing of the remaining files.
+            print(f"::warning::Skipping {ledger_path.name}: {exc}")
+            skipped.append((ledger_path, str(exc)))
+            continue
         if args.check:
             if result.previous != default_branch:
                 mismatches.append(result)
         elif result.changed:
             updated.append(result)
 
+    if skipped:
+        print(f"Skipped {len(skipped)} corrupt ledger(s):")
+        for path, reason in skipped:
+            print(f"  - {path.name}: {reason}")
+
     if args.check:
         if mismatches:
             print("Found ledgers with stale base values:")
diff --git a/templates/consumer-repo/.github/workflows/agents-72-codex-belt-worker.yml b/templates/consumer-repo/.github/workflows/agents-72-codex-belt-worker.yml
index 14e6f2bfb..3695d5226 100644
--- a/templates/consumer-repo/.github/workflows/agents-72-codex-belt-worker.yml
+++ b/templates/consumer-repo/.github/workflows/agents-72-codex-belt-worker.yml
@@ -905,6 +905,7 @@ jobs:
           if gh_output:
               with open(gh_output, 'a', encoding='utf-8') as handle:
                   handle.write(f"task_id={start_info['task']['id'] if start_info['task'] else ''}\n")
+                  handle.write(f"task_title={start_info['task']['title'] if start_info['task'] else ''}\n")
                   handle.write(f"task_status={start_info['task']['current_status'] if start_info['task'] else ''}\n")
                   handle.write(f"ledger_changed={'true' if changed else 'false'}\n")
                   handle.write(f"ledger_created={'true' if start_info['created'] else 'false'}\n")
@@ -1187,12 +1188,28 @@ jobs:
             const prNumber = Number('${{ steps.pr.outputs.number }}');
             const branch = ('${{ steps.ctx.outputs.branch }}' || '').trim() || '(unknown branch)';
             const dryRun = '${{ steps.mode.outputs.dry_run }}' === 'true';
+            const taskId = ('${{ steps.ledger_start.outputs.task_id }}' || '').trim();
+            const taskTitle = ('${{ steps.ledger_start.outputs.task_title }}' || '').trim();
             const { owner, repo } = context.repo;
             const marker = '<!-- codex-activation-marker -->';
             const summary = dryRun
               ? `Codex Worker activated for branch \`${branch}\` (dry run preview).`
               : `Codex Worker activated for branch \`${branch}\`.`;
-            const body = `${marker}\n${summary}\n\n@codex start\n\nAutomated belt worker prepared this PR. Please continue implementing the requested changes.`;
+            // Direct Codex to focus on the single next ledger task for higher
+            // first-commit success probability.  Full issue context is in the
+            // PR body; this comment narrows the immediate scope.
+            let taskDirective = '';
+            if (taskId && taskTitle) {
+              taskDirective = [
+                '',
+                `**Focus on this task first:** \`${taskId}\` — ${taskTitle}`,
+                '',
+                'Implement **only** this task in your first commit.',
+                'Ensure the code compiles and existing tests pass before moving on.',
+                'The keepalive loop will assign subsequent tasks after this one is complete.',
+              ].join('\n');
+            }
+            const body = `${marker}\n${summary}\n\n@codex start${taskDirective}`;
 
             try {
               const comments = await paginateWithRetry(
diff --git a/templates/consumer-repo/.github/workflows/agents-autofix-loop.yml b/templates/consumer-repo/.github/workflows/agents-autofix-loop.yml
index 677da056a..902cdf831 100644
--- a/templates/consumer-repo/.github/workflows/agents-autofix-loop.yml
+++ b/templates/consumer-repo/.github/workflows/agents-autofix-loop.yml
@@ -155,7 +155,7 @@ jobs:
               appendix: '',
               stop_reason: '',
               attempts: '0',
-              max_attempts: '3',
+              max_attempts: '2',
               trigger_reason: 'unknown',
               trigger_job: '',
               trigger_step: '',
@@ -280,7 +280,7 @@ jobs:
             // Reduce attempts for auto-escalated PRs (they weren't agent-initiated)
             const isEscalated = labels.includes('autofix:escalated');
               const maxAttempts = isEscalated
-                ? Math.min(2, Number(outputs.max_attempts))
+                ? 1
                 : Number(outputs.max_attempts);
             const previousRuns = await paginateWithRetry(
               github,
diff --git a/templates/consumer-repo/.github/workflows/agents-pr-meta.yml b/templates/consumer-repo/.github/workflows/agents-pr-meta.yml
index f07a33ee7..24a480544 100644
--- a/templates/consumer-repo/.github/workflows/agents-pr-meta.yml
+++ b/templates/consumer-repo/.github/workflows/agents-pr-meta.yml
@@ -47,8 +47,10 @@ concurrency:
   group: >-
     ${{ github.event_name == 'issue_comment' &&
         format('agents-pr-meta-comment-{0}', github.event.comment.id) ||
+        github.event_name == 'pull_request' &&
+        format('agents-pr-meta-pr-{0}', github.event.pull_request.number) ||
         format('agents-pr-meta-run-{0}', github.run_id) }}
-  cancel-in-progress: false
+  cancel-in-progress: ${{ github.event_name == 'pull_request' }}
 
 jobs:
   # Resolve PR context for issue_comment events
diff --git a/tests/scripts/test_ledger_migrate_base.py b/tests/scripts/test_ledger_migrate_base.py
index 0fe7bac40..89cb4fc9f 100644
--- a/tests/scripts/test_ledger_migrate_base.py
+++ b/tests/scripts/test_ledger_migrate_base.py
@@ -248,10 +248,12 @@ def test_main_check_reports_mismatches(monkeypatch, capsys, tmp_path) -> None:
     agents_dir.mkdir()
     ledger_path = agents_dir / "issue-9-ledger.yml"
     ledger_path.write_text(
-        textwrap.dedent("""\
+        textwrap.dedent(
+            """\
         base: develop
         items: []
-        """),
+        """
+        ),
         encoding="utf-8",
     )
 
@@ -318,3 +320,48 @@ def test_main_reports_no_updates(monkeypatch, capsys, tmp_path) -> None:
     assert exit_code == 0
     out = capsys.readouterr().out
     assert "Ledgers already matched the default branch; no updates written." in out
+
+
+def test_main_skips_corrupt_ledger_and_continues(monkeypatch, capsys, tmp_path) -> None:
+    """A corrupt YAML ledger must not block processing of other ledgers."""
+    agents_dir = tmp_path / ".agents"
+    agents_dir.mkdir()
+    # Write a corrupt ledger with invalid YAML
+    corrupt = agents_dir / "issue-10-ledger.yml"
+    corrupt.write_text("base: main\ntasks:\n  - title: `backtick breaks yaml`\n", encoding="utf-8")
+    # Write a valid ledger that should still be processed
+    valid = agents_dir / "issue-20-ledger.yml"
+    _write_ledger(valid, {"base": "develop", "items": []})
+
+    monkeypatch.setattr(ledger_migrate_base, "find_repo_root", lambda: tmp_path)
+    monkeypatch.setattr(ledger_migrate_base, "detect_default_branch", lambda _=None: "main")
+
+    exit_code = ledger_migrate_base.main([])
+
+    assert exit_code == 0
+    out = capsys.readouterr().out
+    # The corrupt ledger was skipped with a warning
+    assert "Skipping issue-10-ledger.yml" in out
+    assert "Skipped 1 corrupt ledger(s):" in out
+    # The valid ledger was still updated
+    assert yaml.safe_load(valid.read_text(encoding="utf-8"))["base"] == "main"
+
+
+def test_main_check_skips_corrupt_ledger(monkeypatch, capsys, tmp_path) -> None:
+    """--check mode also skips corrupt ledgers without failing."""
+    agents_dir = tmp_path / ".agents"
+    agents_dir.mkdir()
+    corrupt = agents_dir / "issue-5-ledger.yml"
+    corrupt.write_text("not: valid: yaml: `oops`\n", encoding="utf-8")
+    valid = agents_dir / "issue-6-ledger.yml"
+    valid.write_text("base: main\n", encoding="utf-8")
+
+    monkeypatch.setattr(ledger_migrate_base, "find_repo_root", lambda: tmp_path)
+    monkeypatch.setattr(ledger_migrate_base, "detect_default_branch", lambda _=None: "main")
+
+    exit_code = ledger_migrate_base.main(["--check"])
+
+    assert exit_code == 0
+    out = capsys.readouterr().out
+    assert "Skipping issue-5-ledger.yml" in out
+    assert "All ledgers already track the default branch." in out

From 6f5e110beb7aecd8bde358f08b0f2831ae8c307a Mon Sep 17 00:00:00 2001
From: "github-actions[bot]" <github-actions[bot]@users.noreply.github.com>
Date: Thu, 12 Feb 2026 15:31:30 +0000
Subject: [PATCH 07/11] chore(autofix): formatting/lint

---
 tests/scripts/test_ledger_migrate_base.py | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/tests/scripts/test_ledger_migrate_base.py b/tests/scripts/test_ledger_migrate_base.py
index 89cb4fc9f..180eaac5a 100644
--- a/tests/scripts/test_ledger_migrate_base.py
+++ b/tests/scripts/test_ledger_migrate_base.py
@@ -248,12 +248,10 @@ def test_main_check_reports_mismatches(monkeypatch, capsys, tmp_path) -> None:
     agents_dir.mkdir()
     ledger_path = agents_dir / "issue-9-ledger.yml"
     ledger_path.write_text(
-        textwrap.dedent(
-            """\
+        textwrap.dedent("""\
         base: develop
         items: []
-        """
-        ),
+        """),
         encoding="utf-8",
     )
 

From 99b3d5207305a442de453c47cac3a767a91fc22a Mon Sep 17 00:00:00 2001
From: "github-actions[bot]"
 <41898282+github-actions[bot]@users.noreply.github.com>
Date: Thu, 12 Feb 2026 15:35:54 +0000
Subject: [PATCH 08/11] chore(codex-autofix): apply updates (PR #1484)

---
 tests/scripts/test_ledger_migrate_base.py | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/tests/scripts/test_ledger_migrate_base.py b/tests/scripts/test_ledger_migrate_base.py
index 89cb4fc9f..180eaac5a 100644
--- a/tests/scripts/test_ledger_migrate_base.py
+++ b/tests/scripts/test_ledger_migrate_base.py
@@ -248,12 +248,10 @@ def test_main_check_reports_mismatches(monkeypatch, capsys, tmp_path) -> None:
     agents_dir.mkdir()
     ledger_path = agents_dir / "issue-9-ledger.yml"
     ledger_path.write_text(
-        textwrap.dedent(
-            """\
+        textwrap.dedent("""\
         base: develop
         items: []
-        """
-        ),
+        """),
         encoding="utf-8",
     )
 

From 52bd3542ab1c01a4160e9999de5e3219bc53b819 Mon Sep 17 00:00:00 2001
From: "github-actions[bot]"
 <41898282+github-actions[bot]@users.noreply.github.com>
Date: Thu, 12 Feb 2026 15:40:51 +0000
Subject: [PATCH 09/11] chore(codex-autofix): apply updates (PR #1484)

---
 .../scripts/__tests__/detect-changes.test.js  | 62 +++++++++++++++++++
 .github/scripts/detect-changes.js             | 57 ++++++++++++++---
 2 files changed, 110 insertions(+), 9 deletions(-)

diff --git a/.github/scripts/__tests__/detect-changes.test.js b/.github/scripts/__tests__/detect-changes.test.js
index bd7b8c119..09fe2604c 100644
--- a/.github/scripts/__tests__/detect-changes.test.js
+++ b/.github/scripts/__tests__/detect-changes.test.js
@@ -82,3 +82,65 @@ test('detectChanges fetches files via callback', async () => {
   assert.equal(result.outputs.run_core, 'true');
   assert.equal(result.outputs.workflow_changed, 'false');
 });
+
+test('detectChanges falls back to conservative defaults when listFiles is inaccessible', async () => {
+  const warnings = [];
+  const result = await detectChanges({
+    core: {
+      warning(message) {
+        warnings.push(String(message));
+      },
+      setOutput() {},
+    },
+    context: {
+      eventName: 'pull_request',
+      repo: { owner: 'octo', repo: 'demo' },
+      payload: { pull_request: { number: 42 } },
+    },
+    github: {
+      rest: {
+        pulls: {
+          listFiles: async () => ({ data: [] }),
+        },
+      },
+      paginate: {
+        iterator: () => {
+          const error = new Error('Resource not accessible by integration');
+          error.status = 403;
+          throw error;
+        },
+      },
+    },
+  });
+
+  assert.equal(result.outputs.doc_only, 'false');
+  assert.equal(result.outputs.run_core, 'true');
+  assert.equal(result.outputs.reason, 'rate_limited');
+  assert.equal(result.outputs.docker_changed, 'false');
+  assert.equal(result.outputs.workflow_changed, 'true');
+  assert.equal(warnings.length, 1);
+  assert.match(warnings[0], /Unable to determine changed files via API/);
+});
+
+test('detectChanges supports clients without paginate.iterator', async () => {
+  const result = await detectChanges({
+    context: {
+      eventName: 'pull_request',
+      repo: { owner: 'octo', repo: 'demo' },
+      payload: { pull_request: { number: 1 } },
+    },
+    github: {
+      rest: {
+        pulls: {
+          listFiles: async () => ({ data: [] }),
+        },
+      },
+      paginate: async () => [{ filename: 'docs/README.md' }],
+    },
+  });
+
+  assert.equal(result.outputs.doc_only, 'true');
+  assert.equal(result.outputs.run_core, 'false');
+  assert.equal(result.outputs.reason, 'docs_only');
+  assert.equal(result.outputs.workflow_changed, 'false');
+});
diff --git a/.github/scripts/detect-changes.js b/.github/scripts/detect-changes.js
index 45c31c4f5..64d34cb68 100644
--- a/.github/scripts/detect-changes.js
+++ b/.github/scripts/detect-changes.js
@@ -168,6 +168,28 @@ function isRateLimitError(error) {
   return message.includes('rate limit') || message.includes('ratelimit');
 }
 
+function isNonFatalListFilesError(error) {
+  if (!error) {
+    return false;
+  }
+  if (isRateLimitError(error)) {
+    return true;
+  }
+  const status = error.status || error?.response?.status;
+  if ([401, 403, 404, 422].includes(status)) {
+    return true;
+  }
+  const message = String(error.message || error?.response?.data?.message || '').toLowerCase();
+  return (
+    message.includes('resource not accessible by integration') ||
+    message.includes('insufficient permission') ||
+    message.includes('requires higher permissions') ||
+    message.includes('not found') ||
+    message.includes('unprocessable') ||
+    message.includes('validation failed')
+  );
+}
+
 async function listChangedFiles({ github, context }) {
   const pull = context?.payload?.pull_request;
   const number = pull?.number;
@@ -175,25 +197,42 @@ async function listChangedFiles({ github, context }) {
     return [];
   }
   try {
-    const iterator = github.paginate.iterator(github.rest.pulls.listFiles, {
+    const files = [];
+    const params = {
       owner: context.repo.owner,
       repo: context.repo.repo,
       pull_number: number,
       per_page: 100,
-    });
-    const files = [];
-    for await (const page of iterator) {
-      if (Array.isArray(page.data)) {
-        for (const item of page.data) {
+    };
+    if (typeof github?.paginate?.iterator === 'function') {
+      const iterator = github.paginate.iterator(github.rest.pulls.listFiles, params);
+      for await (const page of iterator) {
+        if (Array.isArray(page.data)) {
+          for (const item of page.data) {
+            if (item && typeof item.filename === 'string') {
+              files.push(item.filename);
+            }
+          }
+        }
+      }
+      return files;
+    }
+
+    if (typeof github?.paginate === 'function') {
+      const items = await github.paginate(github.rest.pulls.listFiles, params);
+      if (Array.isArray(items)) {
+        for (const item of items) {
           if (item && typeof item.filename === 'string') {
             files.push(item.filename);
           }
         }
       }
+      return files;
     }
-    return files;
+
+    throw new Error('GitHub paginate API is unavailable');
   } catch (error) {
-    if (isRateLimitError(error)) {
+    if (isNonFatalListFilesError(error)) {
       return null;
     }
     throw error;
@@ -261,7 +300,7 @@ async function detectChanges({ github, context, core, files, fetchFiles } = {})
       workflow_changed: 'true',
     };
     const warn = core?.warning ? core.warning.bind(core) : console.warn.bind(console);
-    warn('Rate limit reached while determining changed files; assuming code changes (but not docker).');
+    warn('Unable to determine changed files via API; assuming code changes (but not docker).');
     if (core) {
       for (const [key, value] of Object.entries(outputs)) {
         core.setOutput(key, value);

From ffda55206332984f4163eba4e20e07c602c0dbc1 Mon Sep 17 00:00:00 2001
From: "github-actions[bot]"
 <41898282+github-actions[bot]@users.noreply.github.com>
Date: Thu, 12 Feb 2026 15:41:08 +0000
Subject: [PATCH 10/11] chore: sync template scripts

---
 .../.github/scripts/detect-changes.js         | 57 ++++++++++++++++---
 1 file changed, 48 insertions(+), 9 deletions(-)

diff --git a/templates/consumer-repo/.github/scripts/detect-changes.js b/templates/consumer-repo/.github/scripts/detect-changes.js
index 45c31c4f5..64d34cb68 100644
--- a/templates/consumer-repo/.github/scripts/detect-changes.js
+++ b/templates/consumer-repo/.github/scripts/detect-changes.js
@@ -168,6 +168,28 @@ function isRateLimitError(error) {
   return message.includes('rate limit') || message.includes('ratelimit');
 }
 
+function isNonFatalListFilesError(error) {
+  if (!error) {
+    return false;
+  }
+  if (isRateLimitError(error)) {
+    return true;
+  }
+  const status = error.status || error?.response?.status;
+  if ([401, 403, 404, 422].includes(status)) {
+    return true;
+  }
+  const message = String(error.message || error?.response?.data?.message || '').toLowerCase();
+  return (
+    message.includes('resource not accessible by integration') ||
+    message.includes('insufficient permission') ||
+    message.includes('requires higher permissions') ||
+    message.includes('not found') ||
+    message.includes('unprocessable') ||
+    message.includes('validation failed')
+  );
+}
+
 async function listChangedFiles({ github, context }) {
   const pull = context?.payload?.pull_request;
   const number = pull?.number;
@@ -175,25 +197,42 @@ async function listChangedFiles({ github, context }) {
     return [];
   }
   try {
-    const iterator = github.paginate.iterator(github.rest.pulls.listFiles, {
+    const files = [];
+    const params = {
       owner: context.repo.owner,
       repo: context.repo.repo,
       pull_number: number,
       per_page: 100,
-    });
-    const files = [];
-    for await (const page of iterator) {
-      if (Array.isArray(page.data)) {
-        for (const item of page.data) {
+    };
+    if (typeof github?.paginate?.iterator === 'function') {
+      const iterator = github.paginate.iterator(github.rest.pulls.listFiles, params);
+      for await (const page of iterator) {
+        if (Array.isArray(page.data)) {
+          for (const item of page.data) {
+            if (item && typeof item.filename === 'string') {
+              files.push(item.filename);
+            }
+          }
+        }
+      }
+      return files;
+    }
+
+    if (typeof github?.paginate === 'function') {
+      const items = await github.paginate(github.rest.pulls.listFiles, params);
+      if (Array.isArray(items)) {
+        for (const item of items) {
           if (item && typeof item.filename === 'string') {
             files.push(item.filename);
           }
         }
       }
+      return files;
     }
-    return files;
+
+    throw new Error('GitHub paginate API is unavailable');
   } catch (error) {
-    if (isRateLimitError(error)) {
+    if (isNonFatalListFilesError(error)) {
       return null;
     }
     throw error;
@@ -261,7 +300,7 @@ async function detectChanges({ github, context, core, files, fetchFiles } = {})
       workflow_changed: 'true',
     };
     const warn = core?.warning ? core.warning.bind(core) : console.warn.bind(console);
-    warn('Rate limit reached while determining changed files; assuming code changes (but not docker).');
+    warn('Unable to determine changed files via API; assuming code changes (but not docker).');
     if (core) {
       for (const [key, value] of Object.entries(outputs)) {
         core.setOutput(key, value);

From e053e6d0728bd3908da64d625e0c3ce10fb4dfde Mon Sep 17 00:00:00 2001
From: stranske <stranske@gmail.com>
Date: Thu, 12 Feb 2026 16:00:49 +0000
Subject: [PATCH 11/11] fix: sanitize task_title for GITHUB_OUTPUT and
 normalize warning annotations

Address inline review feedback on PR #1484:
- Sanitize task_title by replacing newlines/carriage returns with spaces
  before writing to $GITHUB_OUTPUT (prevents broken output parsing)
- Normalize yaml.YAMLError messages to single-line in ::warning:: annotations
  (prevents malformed GitHub Actions annotations)
- Both belt-worker copies updated in sync
---
 .github/workflows/agents-72-codex-belt-worker.yml         | 8 +++++---
 scripts/ledger_migrate_base.py                            | 5 +++--
 .../.github/workflows/agents-72-codex-belt-worker.yml     | 8 +++++---
 3 files changed, 13 insertions(+), 8 deletions(-)

diff --git a/.github/workflows/agents-72-codex-belt-worker.yml b/.github/workflows/agents-72-codex-belt-worker.yml
index 29a4dbefc..15f8271f4 100644
--- a/.github/workflows/agents-72-codex-belt-worker.yml
+++ b/.github/workflows/agents-72-codex-belt-worker.yml
@@ -903,10 +903,12 @@ jobs:
 
           gh_output = os.environ.get('GITHUB_OUTPUT')
           if gh_output:
+              task = start_info['task'] or {}
+              task_title = (task.get('title') or '').replace('\r', ' ').replace('\n', ' ')
               with open(gh_output, 'a', encoding='utf-8') as handle:
-                  handle.write(f"task_id={start_info['task']['id'] if start_info['task'] else ''}\n")
-                  handle.write(f"task_title={start_info['task']['title'] if start_info['task'] else ''}\n")
-                  handle.write(f"task_status={start_info['task']['current_status'] if start_info['task'] else ''}\n")
+                  handle.write(f"task_id={task.get('id', '')}\n")
+                  handle.write(f"task_title={task_title}\n")
+                  handle.write(f"task_status={task.get('current_status', '')}\n")
                   handle.write(f"ledger_changed={'true' if changed else 'false'}\n")
                   handle.write(f"ledger_created={'true' if start_info['created'] else 'false'}\n")
                   handle.write(f"ledger_base_aligned={'true' if base_aligned else 'false'}\n")
diff --git a/scripts/ledger_migrate_base.py b/scripts/ledger_migrate_base.py
index a7fcadc8d..9f87285b0 100644
--- a/scripts/ledger_migrate_base.py
+++ b/scripts/ledger_migrate_base.py
@@ -180,8 +180,9 @@ def main(argv: Iterable[str] | None = None) -> int:
             result = migrate_ledger(ledger_path, default_branch, check=args.check)
         except (MigrationError, yaml.YAMLError) as exc:
             # One corrupt ledger must not block processing of the remaining files.
-            print(f"::warning::Skipping {ledger_path.name}: {exc}")
-            skipped.append((ledger_path, str(exc)))
+            reason = str(exc).replace("\n", " ").replace("\r", " ")
+            print(f"::warning::Skipping {ledger_path.name}: {reason}")
+            skipped.append((ledger_path, reason))
             continue
         if args.check:
             if result.previous != default_branch:
diff --git a/templates/consumer-repo/.github/workflows/agents-72-codex-belt-worker.yml b/templates/consumer-repo/.github/workflows/agents-72-codex-belt-worker.yml
index 3695d5226..ab0801413 100644
--- a/templates/consumer-repo/.github/workflows/agents-72-codex-belt-worker.yml
+++ b/templates/consumer-repo/.github/workflows/agents-72-codex-belt-worker.yml
@@ -903,10 +903,12 @@ jobs:
 
           gh_output = os.environ.get('GITHUB_OUTPUT')
           if gh_output:
+              task = start_info['task'] or {}
+              task_title = (task.get('title') or '').replace('\r', ' ').replace('\n', ' ')
               with open(gh_output, 'a', encoding='utf-8') as handle:
-                  handle.write(f"task_id={start_info['task']['id'] if start_info['task'] else ''}\n")
-                  handle.write(f"task_title={start_info['task']['title'] if start_info['task'] else ''}\n")
-                  handle.write(f"task_status={start_info['task']['current_status'] if start_info['task'] else ''}\n")
+                  handle.write(f"task_id={task.get('id', '')}\n")
+                  handle.write(f"task_title={task_title}\n")
+                  handle.write(f"task_status={task.get('current_status', '')}\n")
                   handle.write(f"ledger_changed={'true' if changed else 'false'}\n")
                   handle.write(f"ledger_created={'true' if start_info['created'] else 'false'}\n")
                   handle.write(f"ledger_base_aligned={'true' if base_aligned else 'false'}\n")