[prompt-analysis] Copilot PR Prompt Analysis - 2026-10-25

[github-actions[bot]]

🤖 Copilot PR Prompt Pattern Analysis - 2026-10-25

Summary

Analysis Period: Last 30 days
Total PRs: 603 | Merged: 506 (83.9%) | Closed: 89 (14.8%) | Open: 8 (1.3%)

Merge rate among completed PRs: 85.0%. No PRs had empty bodies.

Prompt Categories and Success Rates

(Categorized by title keywords; rate = merged / completed.)

Category	Total	Merged	Success Rate
Bug Fix	69	60	88%
Feature Addition	78	64	84%
Refactoring	19	14	74%
Documentation	27	24	89%
Testing	13	11	85%
Other/uncategorized	397	333	85%

Prompt Analysis

✅ Successful Prompt Patterns

• Average prompt length: 164 words (median 102)
• Common keywords: fixes, workflow, coverage, regression, preserve, unchanged, existing, validation
• 100% of merged PR bodies reference files/paths or extensions vs 84% of closed ones (89/89 closed sampled with refs: 14 of 89 → see below)

File/path references: merged 506/506 (100%), closed 14/89 (16%) by regex match on .go/.js/.md/.yml/pkg/. Treat with caution: the regex counts only explicit references; the large gap suggests closed PRs lean on generic/narrative bodies.

Examples (merged):

1. PR Separate workflow title parsing from file-I/O benchmark costs #71927: Separate workflow title parsing from file-I/O benchmark costs → Merged
2. PR Update Pi CLI default to 1.4.0 #71899: Update Pi CLI default to 1.4.0 → Merged
3. PR Detect missing recovery in goroutines launched through local closures #71897: Detect missing recovery in goroutines launched through local closures → Merged

❌ Unsuccessful Prompt Patterns

• Average prompt length: 161 words (median 96) – length is not a differentiator
• Keywords more frequent in closed PRs: document, fallback, failed, compiler, repository
• Broad/multi-area scope and triage-style tasks close more often; refactoring is the weakest category (74%)

Examples (closed):

1. PR Triage setup-script security findings as false positives #71895: Triage setup-script security findings as false positives → Closed
2. PR Handle ignored errors and discarded results #71892: Handle ignored errors and discarded results → Closed
3. PR Refactor long workflow and CLI functions into focused helpers #71890: Refactor long workflow and CLI functions into focused helpers → Closed
4. PR Fix Windows agent runtime setup, dry-run configuration, and audit identities #71882: Fix Windows agent runtime setup, dry-run configuration, and audit identities → Closed (bundles several concerns)

Key Insights

• Specific file references: Merged PRs almost always name concrete files/paths; closed PRs rarely do.
• Length buckets (merge rate): <50 words 82%, 50-150 words 83%, 150-400 words 88%, 400+ words 90% — more detailed prompts do slightly better, so brevity is not the goal.
• Scope: Refactoring (74%) and broad "handle all X" tasks underperform targeted fixes/docs (88-89%).

Recommendations

1. DO: Name specific files/functions and state constraints ("preserve existing behavior", "unchanged").
2. DO: Include regression-test/coverage expectations.
3. AVOID: Bundling multiple unrelated concerns or sweeping refactors in one prompt; split them.

Historical Trends

Date	Merged	Closed	Success Rate
2026-10-25	506	89	85.0%
2026-10-22	165	29	85.1%
2026-10-20	513	101	83.5%
2026-10-15	500	94	84.2%
2026-10-12	472	90	84.0%

Trend: Success rate stable at ~84-85% over the past two weeks.

---

Generated by Copilot PR Prompt Analysis (Run: 40231948970)

Generated by 🔍 Copilot PR Prompt Pattern Analysis · copilot · auto · ~16 AIC · ⌖ ~1 AIC · ⊞ 11.6K · ◷

• expires on Oct 26, 2026, 12:45 PM UTC-08:00

[github-actions[bot]]

This discussion has been marked as outdated by Copilot PR Prompt Pattern Analysis.

A newer discussion is available at Discussion #72178.