[copilot-opt] 56% of workflow runs blocked by action_required approval gate on Copilot branches

[github-actions]

Problem

All workflows triggered from Copilot-authored branches (copilot/*) require manual approval before execution, resulting in a action_required conclusion. This blocks the CI feedback loop: the agent cannot iterate on failures because its own CI cannot run automatically. The copilot/fix-cli-docker-build branch had 13 out of 14 workflow runs blocked in this way — every push triggers a new wave of action_required workflows that accumulate without ever running.

Evidence

• Analysis window: 2026-05-25 to 2026-06-08
• Sessions analyzed: 50
• Key metrics and examples:
  • 28 out of 50 workflow runs (56%) concluded with action_required
  • copilot/fix-cli-docker-build: 13/14 runs were action_required, across CWI, CGO, Smoke CI, Doc Build - Deploy, Content Moderation, AI Moderator, Q, and Agentic Commands
  • copilot/bump-mcp-gateway-to-v0325: 2 CWI runs both action_required
  • copilot/debug-failed-run-27151802516: 1 CWI run action_required
  • Pattern repeats across 4 distinct Copilot branches
  • The approval gate fires immediately on PR open and each subsequent push, requiring manual re-approval

View action_required breakdown by workflow name

Workflow name	action_required count
Agentic Commands	5
CWI	5
Q	5
Smoke CI	4
CGO	3
Doc Build - Deploy	2
AI Moderator	1
Content Moderation	1
Label Closed PRs	1
PR Description Updater	1

Proposed Change

1. Approve the Copilot GitHub App identity (app/copilot-swe-agent) as a trusted actor for PR-triggered workflows, so its branch pushes do not require manual re-approval each time.
2. Alternatively, configure GitHub Actions approval policies to auto-approve runs from copilot/* branches opened by the bot identity.
3. Audit which specific workflow trigger condition is causing the approval gate (e.g., first-time contributor policy vs. fork policy) and address the root policy cause.

Expected Impact

• Eliminate the 56% action_required bottleneck and restore CI-driven iteration for Copilot PRs
• Allow the agent to detect and fix CI failures autonomously without human intervention per push
• Reduce wasted workflow trigger noise from repeated unapproved pushes accumulating on the same branch

Notes

• Distinct root cause category: CI approval gating policy for bot-authored branches
• Data quality: only 1 day of session data available (50 runs); the action_required rate may persist across the full 14-day window given it affects 4 different branches uniformly
• No events.jsonl files were found in session logs; analysis is based on sessions-list.json metadata only

Generated by ⚡ Copilot Opt · ⌖ 44.3 AIC · ⊞ 19.7K · ◷

[github-actions]

This issue is being closed as outdated. A newer issue has been created: #37902

View newer issue

---

This action was performed automatically by the Copilot Opt workflow.