[prompt-analysis] Copilot PR Prompt Analysis - 2026-10-28

[github-actions[bot]]

🤖 Copilot PR Prompt Pattern Analysis - 2026-10-28

Summary

Analysis Period: Last 30 days (2026-09-28 → 2026-10-28)
Total PRs: 606 | Merged: 508 (83.8%) | Closed: 96 (15.8%) | Open: 2 (0.3%)

Merge rate among completed PRs: 84.1%. No PRs had empty bodies.

Prompt Categories and Success Rates

Categorised by title keywords (first match wins; open PRs excluded).

Category	Total	Merged	Success Rate
Bug Fix	69	61	88.4%
Feature Addition	90	76	84.4%
Refactoring	11	9	81.8%
Documentation	23	20	87.0%
Testing	7	5	71.4%
Other	404	337	83.4%

Prompt Analysis

✅ Successful Prompt Patterns

• Average prompt length (merged): 160 words
• Bodies referencing concrete file types (.go/.cjs/.md/.yml/.ts/.js): 16% of merged vs 15% of closed (no meaningful difference)
• Typical merged titles are narrow and specific, e.g. Suppress startup-failure reports for cancelled agent jobs #72405 Suppress startup-failure reports for cancelled agent jobs; Allow on.bots Apps to check out same-repository PRs on edited events #72403 Allow on.bots Apps to check out same-repository PRs on edited events; Honor newer cached action pins in runtime setup #72272 Honor newer cached action pins in runtime setup.

❌ Unsuccessful Prompt Patterns

• Average prompt length (closed): 138 words
• Closed PRs cluster around: duplicate/superseded attempts (Allow on.bots Apps to check out same-repository PR edits #72466 vs merged Allow on.bots Apps to check out same-repository PRs on edited events #72403), [WIP] / "Investigate" / "Triage" / "Assess" tasks ([WIP] Investigate recurring indexing and ignored-error findings #71881, Triage unsubstantiated red-team findings #71621, Assess Sighthound command-injection findings #67990), experimental or design-dependent work ([Experimental] Long-running agentic work across multiple runs #68315, Per-server HTTP MCP CA support requires gateway changes #69436), and broad multi-area hardening (Refactor long workflow and CLI functions into focused helpers #71890 Refactor long workflow and CLI functions, Handle ignored errors and discarded results #71892 Handle ignored errors).

Key Insights

• Length: Success rises with prompt length: <50 words 81.8% (n=11), 50–150 words 83.0% (n=464), 150–300 words 86.6% (n=67), 300+ words 90.3% (n=62). Fuller context helps modestly.
• Task type: Bug fixes and docs merge slightly more often than features/refactors; testing is lowest but n=7 is too small to conclude.
• Investigative/WIP prompts and prompts that duplicate an existing in-flight PR are over-represented among closed PRs.

Recommendations

1. DO: Give detailed context (150+ words: goal, constraints, acceptance criteria) for non-trivial changes.
2. DO: Keep scope to one concrete behaviour change per PR.
3. AVOID: Open-ended "investigate/triage" prompts and re-issuing a task while a prior PR is still open.

Historical Trends

Date	PRs	Success Rate
2026-10-28	606	84.1%
2026-10-27	605	84.1%
2026-10-25	603	85.0%
2026-10-24	585	85.2%
2026-10-23	587	84.1%

Trend: Stable, merge rate within 84–85% over the past week.

---

Generated by Copilot PR Prompt Analysis (Run: 40765121606)

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

• expires on Oct 29, 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 #73039.