[safe-output-health] Safe Output Health Report - 2026-04-13

[github-actions[bot]]

Executive Summary

• Period: 2026-04-13 (last 24 hours)
• Runs Analyzed: 129
• Unique Workflows Active: 67
• Safe Output Items Submitted: 135
• Safe Output Hard Failures: 3
• Non-Fatal Warnings: 1
• Error Clusters: 1
• Overall Success Rate: 97.8% (132/135)
• Health Status: ⚠️ Degraded — recurring rate-limit burst recurred for 3rd time

The primary issue today is the API rate limit concurrent burst pattern — 3 create_issue safe output operations failed with HTTP 403 during a 3-minute window (12:17–12:20 UTC) when multiple daily workflows completed and triggered safe output jobs simultaneously. This is the third occurrence of this pattern (also seen 2026-04-02 and 2026-04-07).

---

Safe Output Job Statistics

Job Type	Total Executions	Failures	Success Rate
create_issue	19	3	84.2%
create_discussion	21	0	100%
add_comment	25	0	100%
create_pull_request	20	0	100%
upload_asset	11	0	100%
noop	11	0	100%
add_labels	10	0	100%
upload_artifact	3	0	100%
update_issue	3	0	100%
assign_to_agent	4	0	100%
submit_pull_request_review	1	0	100%*
dispatch_workflow	1	0	100%
create_pull_request_review_comment	2	0	100%**

*Warning issued but counted as non-failure (see Cluster 2)
**Skipped (not in PR context)

---

Error Clusters

Cluster 1: API Rate Limit Exceeded — create_issue (HIGH severity)

• Count: 3 failures across 2 runs
• Affected Workflows: Workflow Health Manager - Meta-Orchestrator, Multi-Device Docs Tester
• Time Window: 12:17:39–12:20:17 UTC (3-minute burst)
• HTTP Response: POST /repos/github/gh-aw/issues — 403
• Pattern: RECURRING (3rd date: 2026-04-02, 2026-04-07, 2026-04-13)

Error details (run-24342586738 — Workflow Health Manager)

2026-04-13T12:17:39Z POST /repos/github/gh-aw/issues - 403 with id 2430:FD3B5:A5BFD83:2B114625:69DCDEE3 in 96ms
##[error]✗ Failed to create issue "Daily Semgrep Scan: Failure detected Apr 13 2026" in github/gh-aw:
  API rate limit exceeded for installation. Request ID: 2430:FD3B5:A5BFD83:2B114625:69DCDEE3

2026-04-13T12:17:39Z POST /repos/github/gh-aw/issues - 403 with id 2430:3E650:11B2B85:47E47C4:69DCDEE3 in 45ms
##[error]✗ Failed to create issue "Workflow Health Dashboard — Apr 13, 2026" in github/gh-aw:
  API rate limit exceeded for installation. Request ID: 2430:3E650:11B2B85:47E47C4:69DCDEE3

Processing Summary: Total messages: 2, Successful: 0, Failed: 2
```

</details>

<details>
<summary>Error details (run-24342635836 — Multi-Device Docs Tester)</summary>

```
2026-04-13T12:20:17Z POST /repos/github/gh-aw/issues - 403 with id 1680:DFCE9:9CFC4D1:28D76E42:69DCDF81 in 93ms
##[error]✗ Failed to create issue "🔍 Multi-Device Docs Testing Report - 2026-04-13" in github/gh-aw:
  API rate limit exceeded for installation. Request ID: 1680:DFCE9:9CFC4D1:28D76E42:69DCDF81

Processing Summary: Total messages: 1, Successful: 0, Failed: 1
```

</details>

- **Root Cause**: 4 workflows (Workflow Health Manager, Multi-Device Docs Tester, Delight, GitHub MCP Structural Analysis) all started within a 2-minute window (12:10–12:12 UTC) due to the noon daily schedule. Their safe output jobs executed concurrently, exhausting the GitHub App installation rate limit.
- **Impact**: Three issues were not created — daily semgrep scan failure report, workflow health dashboard, and multi-device docs test report for 2026-04-13 are missing from the issue tracker.

#### Cluster 2: PR Review Context Warning — `submit_pull_request_review` (LOW severity)

- **Count**: 1 warning (non-fatal, Failed: 0)
- **Affected Workflow**: Smoke Copilot ([§24344369056](https://github.com/github/gh-aw/actions/runs/24344369056))
- **Event**: `schedule` (not pull_request)

```
##[warning]No review context set - cannot submit review
##[warning]✗ Failed to submit PR review: No review context available
Failed: 0

• Root Cause: submit_pull_request_review was configured but the run was triggered by schedule, not a PR event. The inline create_pull_request_review_comment messages were also skipped (not in PR context). The overall run completed successfully (7/10 messages succeeded).
• Impact: Non-fatal. This is the same intermittent pattern seen 2026-04-04 and 2026-04-06.

---

Root Cause Analysis

API Rate Limit Pattern

The GitHub App installation has a finite hourly request budget. When many workflows complete concurrently (the noon UTC burst), safe output jobs race to make API calls. The safe_output_handler_manager.cjs has zero retry logic for rate-limited responses (retries: 0, retry-exempt-status-codes: 400,401,403,404,422 — notably 403 is exempt). This means rate limit hits are instant, unrecoverable failures.

Historical occurrence frequency:

Date	Failures	Window
2026-04-02	7	12:13–12:14 UTC
2026-04-07	~1	unknown
2026-04-13	3	12:17–12:20 UTC

PR Review Context Warning

submit_pull_request_review handler checks for a PR review context at finalization time. When called via schedule (not PR event), inline comments are skipped and the review context is never set. The handler emits a warning but does not count this as a failure. This is expected behavior for smoke tests that run on schedules — the agent submits review-related operations that only make sense in PR context.

---

Recommendations

Critical Issues (Immediate Action Required)

1. Add retry logic for rate-limited safe output operations

   • Priority: High
   • Root Cause: safe_output_handler_manager.cjs uses actions/github-script with retries: 0 and includes 403 in retry-exempt-status-codes, bypassing all retry logic for rate limit responses.
   • Recommended Action: Implement exponential backoff retry for rate limit 403 responses specifically (distinguishable from permission 403 by error message body). Add 3 retries with 30–60s delays.
   • Affected: Workflow Health Manager, Multi-Device Docs Tester, and any workflow that creates issues during the noon burst.

2. Stagger noon-UTC schedule triggers to reduce burst

   • Priority: Medium
   • Root Cause: Multiple workflows all scheduled at "daily" defaulting to noon UTC create a burst of concurrent completions.
   • Recommended Action: Distribute daily schedules across ±30 minutes from noon (e.g., use cron 15 12 * * *, 30 12 * * *, etc.) for non-time-critical workflows.
   • Affected: All noon-scheduled workflows.

Bug Fixes Required

1. safe_output_handler_manager.cjs — Rate Limit Retry
   • File: actions/safe_output_handler_manager.cjs (or actions/github-script configuration)
   • Problem: Rate limit 403 responses cause immediate, unrecoverable failures with no retry
   • Fix: Detect rate-limit 403 by checking error message for "API rate limit exceeded for installation"; implement up to 3 retry attempts with exponential backoff (30s, 60s, 120s); log each retry attempt with remaining rate-limit headers if available.
   • Affected Jobs: create_issue, add_comment, create_discussion, and all GitHub API write operations.

Configuration Changes

1. Schedule diversification for noon burst workflows
   • Current: Multiple workflows at daily → noon UTC
   • Recommended: Assign explicit staggered cron expressions (e.g., 5 12, 15 12, 25 12) to spread safe output calls over a wider window.
   • Reason: 3 recurrences of rate limit burst demonstrate this is a systemic scheduling issue.

Process Improvements

1. Rate limit monitoring and alerting

   • Current State: Rate limit hits are silent failures; no proactive alerting
   • Proposed: Monitor GitHub API rate limit headers (x-ratelimit-remaining) in safe output jobs; emit a warning annotation when remaining drops below a threshold (e.g., 10% of hourly limit); create a noop or missing_tool report when rate limits are the root cause so the issue is surfaced in the health monitor.

2. PR review context guard

   • Current State: Smoke Copilot emits a non-fatal warning when submit_pull_request_review is called in schedule context
   • Proposed: The agent or workflow config should guard PR-review operations with an event-type check. The create_pull_request_review_comment config could include require_pr_context: true to suppress warnings and cleanly skip when not in PR context.

---

Work Item Plans

Work Item 1: Retry Logic for Rate-Limited Safe Output Operations

• Type: Bug Fix
• Priority: High
• Description: Safe output handler fails immediately on HTTP 403 rate limit responses with no retry. Three instances of data loss (issues not created) over 3 calendar dates.
• Acceptance Criteria:
  • Rate limit 403 responses (identified by message body) trigger retry with exponential backoff
  • Up to 3 retries attempted before final failure
  • Each retry is logged with timestamp and attempt number
  • Non-rate-limit 403 (permission errors) still fail immediately without retry
  • Rate limit recovery succeeds in a test scenario
• Technical Approach: In safe_output_handler_manager.cjs, wrap GitHub API calls with a retry helper that checks error.message.includes('API rate limit exceeded'). Use setTimeout with 30s/60s/120s delays. Consider checking x-ratelimit-reset header for exact reset time.
• Estimated Effort: Small
• Dependencies: None

Work Item 2: Schedule Stagger for Noon-UTC Burst Workflows

• Type: Configuration Enhancement
• Priority: Medium
• Description: Multiple workflows trigger at noon UTC ("daily" default), creating a burst of concurrent API calls in safe output jobs.
• Acceptance Criteria:
  • No more than 3–4 workflows complete safe output jobs in any 5-minute window
  • Noon-UTC workflows are spread across a 30–60 minute window
  • Rate limit burst does not recur on next audit
• Technical Approach: Audit all workflows with schedule: daily and assign explicit staggered cron expressions distributed across 11:45–12:45 UTC. Document the assignment convention to prevent future clumping.
• Estimated Effort: Small
• Dependencies: Work Item 1 (as a backstop even if staggering helps)

---

Historical Context

Daily health trend (last 7 audited days)

Date	Hard Failures	Overall Health	Notes
2026-04-05	0	Excellent	4th zero-failure day
2026-04-06	0	Healthy	1 PR review context warning
2026-04-07	1	Degraded	Rate limit burst (1 failure)
2026-04-08	0	Excellent	100% success
2026-04-09	n/a	—	No audit
2026-04-10	2	Degraded	upload_artifact staging + resolve_review_thread
2026-04-11	1	Degraded	upload_artifact staging missing (day 2)
2026-04-12	n/a	—	No audit
2026-04-13	3	⚠️ Degraded	Rate limit burst (3rd recurrence)

• Error rate trend: Unstable — alternating good/bad days with recurring rate limit as the dominant systemic issue.
• Most reliable job type: add_comment, create_discussion, create_pull_request, upload_asset — all 100% today.
• Most problematic job type: create_issue — all 3 failures today, consistently the job most affected by rate limiting.
• Recurring issue: api_rate_limit_concurrent_burst — 3 dates, 10 total failures. No fix has been implemented despite appearing in recommendations on 2026-04-02.

---

Metrics and KPIs

• Overall Safe Output Success Rate: 97.8% (132/135 operations)
• Hard Failure Rate: 2.2%
• Most Reliable Job Types: add_comment, create_discussion, create_pull_request (100%)
• Most Problematic Job Type: create_issue (84.2%)
• Error Cluster Count: 1 (rate limit)
• New Error Patterns Today: 0 (all known recurring patterns)

---

Next Steps

• [HIGH] Implement retry logic for rate-limited API calls in safe_output_handler_manager.cjs
• [MEDIUM] Stagger noon-UTC schedule triggers across affected workflows to reduce burst load
• [LOW] Add PR review context guard to Smoke Copilot to suppress non-fatal warnings cleanly
• [MONITORING] Track whether rate-limit burst recurs after retry logic is added

References:

• §24342586738 — Workflow Health Manager rate limit failure
• §24342635836 — Multi-Device Docs Tester rate limit failure
• §24344369056 — Smoke Copilot PR review context warning

Generated by Safe Output Health Monitor · ● 594.2K · ◷

• expires on Apr 14, 2026, 1:27 PM UTC

[github-actions[bot]]

This discussion has been marked as outdated by Safe Output Health Monitor.

A newer discussion is available at Discussion #26222.