[prompt-analysis] Copilot PR Prompt Analysis - 2026-06-09

[github-actions[bot]]

Summary

Analysis Period: Last 30 days | Run: §27236307914

Metric	Count	Share
Total PRs	1,000	100%
✅ Merged	787	79.7%
❌ Closed (not merged)	203	20.3%
🔄 Open	10	1.0%

Overall success rate is 79.7%, a slight uptick from 79% seen over Jun 6–8.

Prompt Categories and Success Rates

Category	Total	Merged	Success Rate
📝 Docs	146	121	83%
🧪 Tests	181	151	83%
🔧 Chore/Deps	295	240	81%
✨ Feature	371	295	80%
🔒 Security	137	106	77%
🐛 Bug Fix	400	296	74%
♻️ Refactor	126	91	72%

Note: PRs may match multiple categories; counts reflect overlapping matches.

Prompt Analysis

✅ Successful Prompt Patterns

Common characteristics in merged PRs:

• Average prompt length: 179 words (sweet spot: 100–199 words → 82% success)
• Include code blocks: 86% of merged PRs have fenced code examples vs 76% of closed
• Reference specific files/functions: 83% of merged PRs
• WIP labels: only 1.8% of merged PRs are marked WIP

Top action verbs in merged PRs: change, add, update, test, remove

View Example Successful Prompts

PR #38188 — docs: fix mcp list-tools example to use --server flag → Merged

The CLI docs showed <mcp-server> as a positional argument for mcp list-tools, but --server is a required flag and the only positional argument is an optional workflow name. Following the docs literally silently passes the server name as the workflow filter.

PR #38189 — docs(quick-start): clarify githubnext/agentics origin and simplify .lock.yml paragraph → Merged

Two beginner stumbling blocks on the Quick Start page identified via the Documentation Noob Test: no context on where githubnext/agentics comes from, and a mid-setup deep-dive into the compilation model that interrupts onboarding flow.

PR #38186 — docs: add default runaway-cost guardrails to Cost Management reference → Merged

The Cost Management page now explicitly documents the built-in guardrails that prevent runaway agent spend. It also clarifies how those defaults are overridden via frontmatter and enterprise-level environment variables.

❌ Unsuccessful Prompt Patterns

Common characteristics in closed PRs:

• Average prompt length: 226 words — 26% longer than merged PRs
• WIP mentions: 19.2% of closed PRs contain "WIP" (vs 1.8% merged)
• Very long prompts (400+ words): 45% success rate — lowest of any length bucket
• Over-use of vague verbs: resolve (13% closed vs 2% merged), make (18% vs 6%)

View Example Unsuccessful Prompts

PR #38190 — feat(linters): add execcommand analyzer — flag exec.Command() in context-receiving functions → Closed

109 bare exec.Command calls exist in the codebase vs. 24 exec.CommandContext, with no automated enforcement preventing new ones from landing. This adds a execcommand analyzer (the 24th custom linter)...
(Complex new analyzer — high implementation scope)

PR #38184 — Validate default AI credit guardrails in compiler output, runtime env wiring, and compile logs → Closed

The compiler defaulted max-ai-credits and max-daily-ai-credits through different paths, but the generated runtime wiring did not consistently apply the same fallback order for both metrics...
(Multiple systems touched simultaneously)

PR #38168 — Close 2026-06-09 SPDD spec gaps across ET, AIC, SDK driver, parser, and experiments → Closed

Updates five reviewed specs to resolve daily SPDD findings across ET, AIC, SDK, parser, and experiments...
(Batch operation with too wide a scope)

Key Insights

• Prompt length matters: 100–199 word prompts achieve 82% success. Prompts with 400+ words fall to 45%. Concise, focused descriptions outperform verbose ones.
• Code blocks are a strong positive signal: 86% of merged PRs include fenced code examples. Showing the before/after or exact code context significantly improves clarity.
• WIP is a strong negative signal: PRs marked WIP have a 19× higher rate of being closed without merging. If a PR isn't ready, it shouldn't be opened.
• Docs and tests have the highest success rates (both 83%). These tasks are well-defined, bounded, and verifiable — qualities that transfer well to Copilot prompts.
• Refactoring underperforms (72%): Large-scope structural changes require more context and have higher review friction. Breaking refactors into smaller targeted PRs improves outcomes.

Recommendations

Based on today's analysis:

1. DO keep prompts between 100–200 words — this is the optimal range with 82% merge rate
2. DO include code blocks with before/after examples or error output to ground the change
3. DO frame bug-fix prompts with reproduction steps or error messages (not just "fix X")
4. AVOID opening PRs with "WIP" in the title or body — these close 19× more often
5. AVOID prompts over 400 words or spanning many unrelated files/systems simultaneously
6. AVOID vague verbs like "resolve", "make", "check" without specific targets — prefer "add", "update", "remove" with precise scopes

Historical Trends

Date	PRs	Merged	Success Rate
2026-06-09 (today)	1,000	787	79.7%
2026-06-08	1,000	794	79.0%
2026-06-07	1,000	791	79.0%
2026-06-06	1,000	788	79.0%
2026-06-02	1,000	809	80.0%
2026-06-01	1,000	802	80.0%
2026-05-30	1,000	802	80.0%

Trend: After a dip to 79% during Jun 6–8, today's rate ticked back up to 79.7%. The peak of 80.5% was observed in mid-May. The data suggests a mild regression in prompt quality over the past two weeks.

---

References:

• §27236307914

Generated by 🔍 Copilot PR Prompt Pattern Analysis · ⌖ 27.2 AIC · ⊞ 20.7K · ◷

• expires on Jun 10, 2026, 1:29 PM UTC-08:00