[prompt-analysis] Copilot PR Prompt Analysis - 2026-04-30

[github-actions[bot]]

Summary

Analysis Period: Last 30 days
Total PRs: 1,000 | Merged: 775 (77.8%) | Closed (not merged): 221 (22.1%) | Open: 4 (0.4%)

This is the first run of the Copilot PR Prompt Pattern Analysis, establishing a baseline for future trend tracking.

Prompt Categories and Success Rates

Category	Total	Merged	Closed	Success Rate
Bug Fix	712	556	153	78.4%
Feature Addition	253	191	61	75.8%
Testing	13	11	2	84.6%
Other	11	9	2	81.8%
Refactoring	4	3	1	75.0%
Documentation	7	5	2	71.4%

Commit Type Success Rates

Prefix	Merged	Total	Success Rate
ci:	5	5	100%
test:	7	7	100%
docs:	23	27	85.2%
fix:	186	231	80.5%
feat:	76	95	80.0%
perf:	13	16	81.2%
chore:	14	18	77.8%
refactor:	17	22	77.3%
No prefix	434	575	75.5%

Prompt Analysis

✅ Successful Prompt Patterns

Common characteristics in merged PRs:

• Average prompt body length: ~1,161 words
• Median body length: ~333 words
• Titles follow conventional commit format (fix:, feat:, docs:, etc.)
• Prompts typically describe the problem before the solution
• Reference specific files, functions, or CI job names

Example successful prompts:

1. PR fix: remove unused name field in listFieldCase to fix lint-go CI #29239 (fix:): fix: remove unused name field in listFieldCase to fix lint-go CI — described exact CI failure, specific field name → Merged
2. PR feat: parameterize safe-output boolean controls for reusable workflows #29235 (feat:): feat: parameterize safe-output boolean controls for reusable workflows — described limitation and need → Merged
3. PR feat: parameterize safe-output PR policy fields in workflow_call workflows #29230 (feat:): feat: parameterize safe-output PR policy fields in workflow_call workflows — concrete use case with before/after context → Merged

❌ Unsuccessful Prompt Patterns

Common characteristics in closed PRs:

• Average prompt body length: ~1,058 words (slightly shorter)
• Median body length: ~333 words (similar)
• Higher frequency of engine-specific keywords (codex, claude) in closed PRs
• General "improvement" prompts without specific error context

Example closed prompts:

1. PR Enable requireCleanGit in skill optimizer config #29226: Enable requireCleanGit in skill optimizer config — configuration change without sufficient context → Closed
2. PR fix(security): replace curl-pipe-bash with actions/setup-cli (RGS-006) #29209 (fix(security):): Replace curl-pipe-bash pattern — security fix with good intent but potentially too broad in scope → Closed
3. PR feat: add shared/gh-skill.md workflow for gh skill install #29192 (feat:): add shared/gh-skill.md workflow — new workflow addition → Closed

Key Insights

• CI and test-related PRs have perfect merge rates (100%): Prompts that target CI infrastructure or test coverage appear to have clear, measurable success criteria that reviewers can easily verify.
• Prompts referencing specific technical artifacts succeed more: The most successful merged PRs reference exact file names, function names, CI job names, or error messages — providing concrete context.
• Engine-specific PRs (codex, claude) have lower success rates: PRs mentioning specific AI engines in their titles (codex: 14 merged vs 17 closed; claude: 9 merged vs 11 closed) perform slightly below average, suggesting engine-configuration changes face more scrutiny.
• Prompt length matters less than specificity: Merged and closed PRs have similar median body lengths (~333 words), indicating that quality of context outweighs quantity.

Recommendations

Based on today's analysis:

1. DO: Use conventional commit prefixes (fix:, feat:, ci:, test:) — PRs with these prefixes average 77–100% success vs 75.5% for untyped titles
2. DO: Include specific error messages, CI job names, or file/function references in the prompt — these are hallmarks of successful PRs
3. DO: Focus on test and CI improvements — 100% merge rate suggests these are well-received change types
4. AVOID: Vague "improvement" or "enhancement" descriptions without concrete before/after context
5. AVOID: Broad scope changes that touch many unrelated areas — engine-configuration PRs and multi-area refactors show lower success rates

Historical Trends

First run — baseline established. Historical trend data will appear in future reports.

Date	PRs	Success Rate	Top Category
2026-04-30	1,000	77.8%	Bug Fix (71.2% of PRs)

---

References: §25148750109

Generated by Copilot PR Prompt Pattern Analysis · ● 455.1K · ◷

• expires on May 1, 2026, 5:22 AM UTC