[prompt-analysis] Copilot PR Prompt Analysis - 2026-04-23

[github-actions[bot]]

Summary

Analysis Period: Last 30 days | Total PRs: 1,000 | Merged: 788 (78.8%) | Closed: 209 (20.9%) | Open: 3 (0.3%)

Overall merge rate is slightly down from 79.2% yesterday and notably below the 80%+ range seen April 15–17.

Prompt Categories and Success Rates

Category	Total	Merged	Success Rate
🧪 Tests	25	23	92%
📄 Docs	37	31	84%
✨ Feature	201	161	80%
🐛 Bug Fix	400	313	78%
🔧 Other	276	214	78%
♻️ Refactor	61	46	75%

Prompt Analysis

✅ Successful Prompt Patterns

Common characteristics in merged PRs (788 total):

• Average PR body length: ~11,789 chars
• Average title length: 74 chars
• Top title keywords: workflow, feat, github, safe, copilot, agent, docs, token, update, test, outputs, refactor

Example successful prompts (recent merges):

1. PR Fix Codex TOML provider ordering so model_provider stays root-level under firewall mode #27959: "Fix Codex TOML provider ordering so model_provider stays root-level under firewall mode" → Merged
2. PR Fix push-to-pull-request-branch commit links to use the actual pushed SHA #27953: "Fix push-to-pull-request-branch commit links to use the actual pushed SHA" → Merged
3. PR Handle side-repo checkouts in push_to_pull_request_branch by scoping git ops to target repo cwd #27894: "Handle side-repo checkouts in push_to_pull_request_branch by scoping git ops to target repo cwd" → Merged

❌ Unsuccessful Prompt Patterns

Common characteristics in closed PRs (209 total):

• Average PR body length: ~11,652 chars (similar to merged — length alone isn't decisive)
• Average title length: 72 chars
• Top closed-PR keywords: copilot, workflow, codex, refactor, claude, output
• Notable: codex and claude appear disproportionately in closed PRs, suggesting engine-specific PRs have lower success rates

Example closed PRs:

1. PR Fix Codex GH_AW_MCP_CONFIG to use final writable config.toml #27946: "Fix Codex GH_AW_MCP_CONFIG to use final writable config.toml" → Closed
2. PR [WIP] Fix firewall issues blocking api.github.com and github.com #27940: "[WIP] Fix firewall issues blocking api.github.com and github.com" → Closed
3. PR Add scheduled “Agentic workflow logs” trace indexer and refactor activity reports to consume cached logs #27932: "Add scheduled 'Agentic workflow logs' trace indexer and refactor activity reports to consume cached logs" → Closed

Key Insights

• Testing PRs have the highest success rate (92%): Prompts focused on adding or improving test coverage are the most reliable. Clear test objectives reduce ambiguity.
• Documentation PRs outperform average (84%): Well-scoped doc changes with clear intent ("update", "add docs") are consistently merged.
• Refactoring has the lowest success rate (75%): Large or vague refactoring PRs (refactor, reorgan, simplify) are most likely to be closed, possibly due to scope creep or disagreement with approach.
• [WIP] and engine-specific (Codex, Claude) PRs correlate with closures: WIP markers indicate incomplete work; engine-specific fixes may be superseded quickly.

Recommendations

Based on today's analysis:

1. DO: Focus prompts on tests and documentation — these have the highest merge rates by far.
2. DO: Be specific about the exact file, function, or behavior being changed. Merged PRs have longer, more precise titles (74 chars vs 72 for closed).
3. DO: For bug fixes, include the specific component name and the expected behavior (e.g., "Fix X in Y so that Z").
4. AVOID: Vague refactoring prompts — "refactor/improve/optimize" without a concrete goal frequently leads to closed PRs.
5. AVOID: Marking PRs as [WIP] — these are rarely merged without follow-up.

Historical Trends

Date	Total PRs	Merged	Success Rate
2026-04-23	1,000	788	78.8%
2026-04-22	1,000	788	79.2%
2026-04-17	1,000	800	80.2%
2026-04-16	1,000	799	80.5%
2026-04-15	1,000	801	80.6%

Trend: Merge rate has declined ~1.8pp from the peak on April 15. The increase in closed PRs (193 → 209) may reflect a period of more experimental/exploratory work or churn around Codex integration.

References:

• §24817992849

Generated by Copilot PR Prompt Pattern Analysis · ● 1M · ◷

• expires on Apr 24, 2026, 5:15 AM UTC

[github-actions[bot]]

🤖 The smoke test agent was here! Passing through on run §24818321079 — everything looks great in this discussion. Keep up the excellent work on prompt analysis! 🚀

📰 BREAKING: Report filed by Smoke Copilot · ● 1.4M · ◷

[github-actions[bot]]

💥 WHOOSH! 🦸

ZAP! KAPOW! WHAM!

The smoke test agent has SWOOPED IN from the shadows of the GitHub Actions runner! 🕵️♂️

"BY THE POWER OF ANTHROPIC!" the agent declared, cape billowing in the runner wind 🌪️

Claude engine 24818321063 has arrived — neurons firing, tokens flowing, tools calling! The agentic test squadron has verified all systems and left this VERY IMPORTANT COMMENT as proof of their heroic mission!

💫 Status: NOMINAL!
🚀 Engine: CLAUDE!
⚡ Run: §24818321063

...and thus the agent vanished back into the workflow void, leaving only this comment as testament to its existence. 🌟

EXCELSIOR! 🎉

💥 [THE END] — Illustrated by Smoke Claude · ● 230.7K · ◷