[copilot-session-insights] Daily Copilot Agent Session Analysis — 2026-06-02

[github-actions[bot]]

🤖 Copilot Agent Session Analysis — 2026-06-02

Data-quality note: Conversation transcript logs were unavailable (0 files in logs/) — the 10th+ consecutive day of the conversation_log_fetch_failure pattern. This is metadata-only analysis: session outcomes, durations, and branch concentration — not the agent's internal reasoning.

Executive Summary

• Sessions Analyzed: 50 (GitHub Actions workflow runs), 2026-06-02 06:52–08:00 UTC (~68-min burst)
• Completion Rate: 16% (8/50) — up from 8% yesterday, but below the 7-day avg of 24.6%
• Average Duration: 1.52 min (median 0 min — see note)
• Dominant mode: 78% (39/50) ended action_required at zero duration — permission/approval gating, not agent failure
• Experimental Strategy: None (standard analysis)

Key Metrics

Metric	Value	Trend
Total Sessions	50	→
Successful Completions	8 (16%)	↑ vs 06-01 (4); ↓ vs 7-day avg
Failed	3 (6%)	→
Gated / Never-ran (action_required)	39 (78%)	→ vs 06-01 (41, 82%)
Average Duration	1.52 min	↑ vs 06-01 (1.11)
Long sessions ≥15 min	1 (2%)	↓ vs 06-01 (2)

Median = 0 artifact: with 39/50 zero-duration gated runs, median duration is 0. The 11 executed sessions ran 18 s – 19 min.

Session Composition

All 50 sessions ran on just 3 branches: copilot/create-github-agentic-workflow (19, 38%), copilot/duplicate-code-permission-denied (17, 34%), copilot/update-effective-tokens-flow (14, 28%).

• 8 successes (18 s – 19 min): Running Copilot Code Review, Running Copilot cloud agent, Smoke CI, "Addressing comment on PR Add on-demand token audit workflow with date-range dispatch input #36412".
• 3 failures: all the CGO workflow (~3 min each) — a single recurring failing gate.

Success Factors ✅

1. Agent-owned branch + sustained iteration — all 8 successes sit on branches carrying real PR work (concentrated_branch_activity pattern).
2. Executed runs tend to finish — 8 of 11 non-gated runs succeeded (73%). Once a run clears gating, it usually completes.
3. Long sessions = productive work — the lone ≥15-min run (19 min) was a successful agent/review run (long_session_clustering).

Failure Signals ⚠️

1. Permission gating dominates (78%) — 39/50 never executed (action_required, zero duration). Largest source of "non-completion"; it's scheduling friction, not agent failure.
2. Chronically-failing CGO gate — all 3 hard failures from one workflow → persistent gate/config defect.
3. No transcript visibility — 10+ days without logs means loops, context confusion, and recovery quality cannot be measured.

Orphaned Branch Escalation Alerts 🚨

• Orphaned Branches Today: 0 of 3 active branches (0%) — ✅ NORMAL (baseline ~40%, threshold 50%).
• Only 1 in-progress run repo-wide at scan (this workflow, on main); max simultaneous gates on any branch = 1, far below the ≥5 trigger.
• Of 4 open PRs: 2 Copilot-assigned (Recompute effective tokens from raw usage with current weights/multipliers #36421, Refactor harness permission-denied handling into shared helper module #36415), 2 automation PRs with no gate sweep ([actions] Update GitHub Actions versions - 2026-06-02 #36426, [jsweep] Clean validate_memory_files.cjs #36404).
• CI waste from orphans: 0 today. (The 78% gating cost is separate — gated runs queue for approval rather than burning CI.)

📈 Session Trends Analysis

Completion Patterns

Completion rebounded to 16% from yesterday's 8%, driven by 8 successes vs 4 — but stays below the 7-day avg (24.6%) and far from the window peaks of 44–46% (05-23, 05-26). The 14-day series is highly oscillatory (0% on 05-25; a 24-failure spike on 05-31), reflecting gating and a tiny active-branch set rather than steady throughput.

Duration & Efficiency

Average duration edged up to 1.52 min but stays in the low, gating-dominated band; the ≥15-min "productive" cluster shrank to 1 (from 2). Window peaks in long sessions (05-23: 12; 05-26/05-28: 7 each) align with the high-completion days — long-session presence remains the clearest signal of real agent work.

Actionable Recommendations

For users: (1) Pre-authorize routine gated workflows — 78% stall at action_required; grant standing approval for low-risk automation (label/PR-description/moderation) so they run. (2) Keep work on a single agent-owned branch — that's where successes cluster.

For system: (1) Fix conversation-log ingestion (HIGH) — 10+ days of missing transcripts blocks all behavioral analysis. (2) Split "gated" from "failed" in metrics (MED) — reporting action_required as non-completion hides the true 73% executed-success rate. (3) Fix the recurring CGO failure (MED).

For tools: (1) Transcript-fetch health-check/retry (needed every run since 2026-05-24). (2) Gating-aware scheduling to batch or auto-approve perpetually-blocked workflows.

Statistical Summary & Diagnostics

Total Sessions:          50
Successful:              8  (16%)
Failed:                  3  (6%)   [all CGO]
Gated / Never-ran:       39 (78%)  [action_required, zero-duration]
In-Progress:             0

Avg Duration:            1.52 min
Median Duration:         0.00 min  (39/50 zero-duration)
Longest:                 ~19 min (1138 s, success)
Shortest executed:       18 s
Long (>=15 min):         1  (2%)
Tool/Gate Failures:      3  (CGO)

Branches:                3 unique, top share 38%
Orphaned branches:       0  (baseline ~40%)
Conversation logs:       UNAVAILABLE (10+ consecutive days)

Gated workflows (stalled at approval): AI Moderator, Agentic Commands, CGO, CJS, Content Moderation, Design Decision Gate, Doc Build-Deploy, Label Closed PRs, Matt Pocock Skills Reviewer, PR Code Quality Reviewer, PR Description Updater, Q, Smoke CI, Test Quality Sentinel. Loop and clarification-request detection are unavailable without transcripts.

Next Steps

• Restore conversation-log ingestion to re-enable behavioral analysis
• Add a "gated / never-ran" bucket to completion dashboards
• Investigate the recurring CGO workflow failure
• Review auto-approval policy for low-risk gated automation

References: §26807140963 (this run) · §26805936210 (Code Review, success) · §26806671685 (PR Description Updater, gated)

---

Generated 2026-06-02 · Run ID 26807140963 · Workflow: Copilot Session Insights

Generated by 📊 Copilot Session Insights · opus48 2.3M · ◷

• expires on Jun 3, 2026, 8:44 AM UTC

[github-actions[bot]]

This discussion has been marked as outdated by Copilot Session Insights.

A newer discussion is available at Discussion #36596.