[prompt-analysis] Copilot PR Prompt Analysis - 2026-05-07

[github-actions[bot]]

Summary

Analysis Period: Last 30 days (up to 1,000 most recent PRs from copilot/* branches)
Total PRs: 1,000 | Merged: 777 (77.7%) | Closed: 213 (21.3%) | Open: 10 (1.0%)

Overall, Copilot PRs continue to achieve a strong 77.7% merge rate — consistent with the 7-day historical trend (77.7–78.2%). The most significant predictor of failure identified this run: [WIP] prefix correlates with 92% close rate (35 closed vs 3 merged).

Prompt Categories and Success Rates

Category	Total	Merged	Success Rate
Test	7	7	100%
Docs	13	12	92%
Refactor	82	67	82%
Bug Fix	434	341	79%
Feature Addition	205	151	74%
Other	259	199	77%

Prompt Analysis

✅ Successful Prompt Patterns

Common characteristics in merged PRs:

• Average PR title length: ~9.6 words
• Titles with backtick-quoted code references (e.g., `pkg/workflow`) achieve 85% merge rate
• Conventional commit prefixes refactor:, docs:, test:, perf: all exceed 86% merge rate
• Non-WIP (completed) PRs: 81% merge rate

Example successful prompts:

1. PR #30895: fix: CLI consistency - standardize write+ terminology, mcp list-tools example → Merged — focused, specific, references exact commands
2. PR #30894: refactor(agentdrain): rename match and firstKey to descriptive names → Merged — scoped to package, names exact functions
3. PR #30881: Fix lint-go failure by replacing static fmt.Errorf usages flagged by perfsprint → Merged — root cause + fix + tool reference

❌ Unsuccessful Prompt Patterns

Common characteristics in closed PRs:

• [WIP] prefix: 35 closed, only 3 merged (8% merge rate) — these are abandoned work-in-progress branches
• chore prefix: 68% merge rate, lowest of all conventional commit prefixes
• Prompts framing broad investigations without clear scope (e.g., "investigate and optimize...")
• Prompts targeting volatile / rapidly-changing areas (AWF config version specifics)

Example unsuccessful prompts:

1. PR #30896: Migrate 14 workflows to shared/repo-memory-standard.md; add file-glob override support → Closed — large scope, superseded by other work
2. PR #30768: [WIP] Investigate and optimize API calls for Smoke suite → Closed — WIP investigation without concrete fix
3. PR #30769: Fix: add version gate for apiProxy.models in AWF config (AWF v0.25.38+) → Closed — version-specific guard superseded by dependency changes

Key Insights

• WIP is a strong failure signal: [WIP] PRs close at 92% — either because the investigation concludes "no change needed" or the work gets subsumed by another PR. Prefer creating PRs only when changes are ready.
• Scope precision drives success: The top-performing categories (test, docs, refactor) all share a narrow, well-defined scope. Bug fix PRs that name specific files/functions/linters succeed at 81% vs broader ones.
• Conventional commit prefixes help: PRs using refactor:, docs:, test:, perf:, feat: prefixes outperform bare-title PRs (74% for no-prefix). The chore prefix underperforms at 68%.

Recommendations

Based on today's analysis:

1. DO: Use conventional commit prefixes — especially refactor:, docs:, test: — to signal clearly scoped changes (all >85% merge rate)
2. DO: Reference specific files, functions, or tool names in the PR title (backtick-quoted identifiers correlate with 85% merge rate vs 74% for plain text)
3. DO: Include root cause + fix description in PR body — merged PRs average 64 body-words of focused description before the quoted issue section
4. AVOID: Creating PRs with [WIP] prefix unless the goal is to signal "investigation in progress" — they close 92% of the time
5. AVOID: Broad investigation prompts ("investigate and optimize X") — these produce [WIP] PRs that rarely land

Historical Trends

Date	PRs	Merge Rate	Top Category
2026-05-07	1,000	77.7%	bug_fix (434)
2026-05-06	1,000	78.0%	bug_fix (636)
2026-05-04	1,000	78.0%	bug_fix (598)
2026-04-30	1,000	77.8%	bug_fix (884)
2026-04-27	1,000	78.0%	bug_fix (784)
2026-04-26	1,000	78.1%	bug_fix (893)
2026-04-25	1,000	78.2%	bug_fix (780)

Trend: Merge rate has been remarkably stable in the 77.7–78.2% band over the past 12+ days. The slight dip to 77.7% today is within normal variance. Bug Fix remains the dominant category by volume, though it has become more balanced with Feature and Refactor prompts compared to late April when bug_fix represented 80–89% of all PRs.

References:

• Workflow Run §25521679271

Generated by Copilot PR Prompt Pattern Analysis · ● 8.1M · ◷

• expires on May 8, 2026, 9:04 PM UTC