[prompt-analysis] Copilot PR Prompt Analysis - 2026-05-21

[github-actions[bot]]

🤖 Copilot PR Prompt Pattern Analysis - 2026-05-21

Summary

Analysis Period: Last 30 days
Total PRs Analyzed: 993 | Merged: 799 (80.5%) | Closed: 194 (19.5%)

✅ Success rate is 80.5% - slightly above the 7-day average of 78.6%

Key Findings

Prompt Characteristics Comparison

Metric	Merged PRs	Closed PRs	Insight
Average Length	963 words	1105 words	✅ Shorter is better
With Code Blocks	81.7%	76.3%	✅ Code examples help
With File Refs	91.9%	91.2%	➖ Standard practice

Category Success Rates

Category	Total PRs	Merged	Success Rate
🟢 Bug Fix	852	696	81.7%
🟢 Update	886	719	81.2%
🟢 Testing	788	639	81.1%
🟡 Feature	915	736	80.4%
🟡 Documentation	592	470	79.4%
🟡 Refactor	273	214	78.4%
🟡 Removal	316	246	77.8%
🔴 Other	3	1	33.3%

Detailed Analysis

✅ What Makes Successful Prompts?

Common characteristics in merged PRs:

1. Concise and focused: Average 963 words (median: 25 words)
2. Include code examples: 81.7% include code blocks showing before/after or examples
3. Reference specific files: 91.9% mention specific file paths or function names
4. Clear problem statement: Explain what's broken or needed, not just what to do

Example successful prompts:

1. PR #33841: Fix Codex lock compiler: move --model flag after exec subcommand

   The Codex lock compiler was emitting the --model flag before the exec subcommand. Codex CLI treats flags in that position as global options and ...

2. PR #33836: Apply progressive disclosure to ready_for_review PR comment workflows

   PR comments from ready_for_review-triggered workflows (Test Quality Sentinel, Design Decision Gate, Matt Pocock Skills Reviewer, PR Code Quality Rev...

3. PR #33831: Normalize context cancellation checks to errors.Is in CLI paths

   Three production call sites compared Context.Err() to context.Canceled / context.DeadlineExceeded with ==, while the rest of the codebase alre...

❌ Patterns in Closed PRs

Common characteristics:

1. Longer prompts: Average 1105 words - may indicate over-specification or unclear goals
2. Lower code example usage: Only 76.3% include code blocks
3. Task lists without context: Some use checklists without explaining the underlying problem

Example unsuccessful prompts:

1. PR #33839: [WIP] Emit langfuse.session.id and langfuse.user.id for views

   • Add langfuse.session.id and session.id in buildEpisodeAttributesFromContext - [ ] Add langfuse.user.id and user.id in `sendJobSetupSpa...

2. PR #33817: [WIP] Refactor to extract progressive disclosure guidelines into shared component

   Thanks for asking me to work on this. I will get started on it and keep this PR's description up to date as I form a plan and make progress. > ---- >...

3. PR #33815: compiler: add parallel file compilation with --workers n

   The compiler processes workflow files sequentially; with large repos this is a bottleneck. This adds a --workers n flag (default max(1, nCPU-1)) t...

Key Insights

1. Shorter prompts perform better: Merged PRs have prompts averaging 963 words, while closed PRs average 1105 words (142 words longer). Conciseness appears to correlate with success.

2. Code examples help: 81.7% of merged PRs include code blocks vs 76.3% of closed PRs. Including specific code examples increases success rate by 5.4%.

3. File references are standard: 91.9% of both successful and unsuccessful PRs reference specific files. This is a baseline practice, not a differentiator.

4. Category performance: Bug Fix prompts have the highest success rate at 81.7%, while Other prompts have 33.3%.

Recommendations

Based on today's analysis of 993 PRs:

✅ DO:

1. Keep it concise: Aim for prompts under 100 words when possible. Focus on the problem and key constraints.
2. Include code examples: Show before/after snippets or provide example code for context.
3. State the problem clearly: Explain why something needs to change, not just what to change.
4. Reference specific locations: Mention file paths, function names, or line numbers.
5. Focus on bug fixes and features: These categories have 81.7% and 80.4% success rates respectively.

❌ AVOID:

1. Over-specifying implementation: Don't write the code in the prompt - let Copilot determine the best approach.
2. Bare task lists: Checklists without context ("Add X to Y") lack the reasoning needed for good solutions.
3. Vague descriptions: "Improve X" or "Fix Y" without explaining the problem or expected behavior.
4. Multiple unrelated changes: Keep prompts focused on a single problem or feature.

💡 Template for High-Success Prompts:

[Brief problem statement - 1-2 sentences]

[Current behavior or issue]

[Expected behavior or desired outcome]

[Optional: Code example or specific file references]

[Optional: Constraints or important considerations]

Historical Trends (7-Day View)

Date	Total PRs	Success Rate	Top Category	Trend
2026-04-30	1000	77.8%	Other	➡️
2026-05-04	1000	78.0%	Feature	📈
2026-05-06	1000	78.0%	Test	➡️
2026-05-07	1000	77.7%	Test	📉
2026-05-11	1000	79.1%	Refactor	📈
2026-05-12	1000	79.3%	Refactor	📈
2026-05-20	1000	80.6%	Bug Fix	📈

Trend: Success rate has improved by 2.8% over the past week.

---

Analysis of 993 Copilot PRs from the last 30 days • Generated: 2026-05-21 21:16 UTC

Generated by 🔍 Copilot PR Prompt Pattern Analysis · ● 1.2M · ◷

• expires on May 22, 2026, 9:19 PM UTC

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-05-22T21:19:43.571Z.

Closed by Workflow