[prompt-analysis] Copilot PR Prompt Analysis - 2026-04-22 #27774
Replies: 3 comments
-
|
🎉 Smoke test agent reporting in from run §24762218449! I reviewed container digest pinning, built the binary, composed haikus, and ran 14 tests — all green! 🟢 Digests pinned at dawn, See you next run! 🤖
|
Beta Was this translation helpful? Give feedback.
-
|
💥 WHOOSH! ⚡ KA-BOOM! 🦸 The Claude Smoke Test Agent swoops in from the digital cosmos... "By the power of agentic workflows — I WAS HERE!" 💫
...and with a final ZZZAP! the agent vanishes back into the workflow matrix. 🌟 "With great agentic power comes great smoke-tested responsibility." 🌟
|
Beta Was this translation helpful? Give feedback.
-
|
This discussion has been marked as outdated by Copilot PR Prompt Pattern Analysis. A newer discussion is available at Discussion #27975. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Summary
Analysis Period: Last 30 days | Run: §24761281623
Prompt Categories and Success Rates
Bug fixes dominate the dataset (60.6% of all PRs), reflecting this project's active development pace.
✅ Successful Prompt Patterns
Common characteristics in merged PRs:
[WIP]tag: Non-WIP PRs have an 83.5% success rate vs 1.9% for WIP PRsci:andtest:prefixes achieve 100% merge rateConventional commit prefix success rates:
ci:test:deps:docs:feat:fix:refactor:View merged PR examples
Example 1 — #27751:
Prevent built-in jobs.* customization entries from creating agent needs duplicatesExample 2 — #27742:
Fix firewall cleanup permissions to include audit logs directoryExample 3 — #27745:
ci: include runner-guard in cgo security-scan matrix❌ Unsuccessful Prompt Patterns
Common characteristics in closed PRs:
[WIP]tag: 25.1% of closed PRs are tagged[WIP]— these almost never merge (1.9% success)View closed PR examples
Example 1 — #27695 (WIP):
[WIP] Cap Documentation Unbloat at max_turns:30 and require at least 1 safe outputExample 2 — #27672 (WIP):
[WIP] Downgrade Agent Persona Explorer to claude-haiku-4-5 and optimize data fetchingExample 3 — #27744 (non-WIP, closed):
Apply container digest pinning during gh aw updateKey Insights
🚨 WIP = Almost Certain Closure: PRs tagged
[WIP]close 98% of the time. If a Copilot PR starts with boilerplate ("Thanks for asking me to work on this...") with no concrete implementation details, it's almost always destined to be closed.📏 Optimal Prompt Length is 150–299 Words: This range achieves 85% success. Very short (<50 words: 70%) and medium-long (300–499 words: 71%) prompts underperform. Prompts over 500 words still succeed at 78% due to the long setup/context included in standard Copilot PR templates.
🧪 Testing PRs Are Safest: With a 96% success rate, test-focused prompts almost always merge — likely because they have clear, verifiable goals with no architectural disagreements.
Recommendations
Based on today's analysis:
fix:,feat:,ci:,test:) — they signal clear intent and correlate with higher merge rates[WIP]state — they almost never merge (1.9% rate); if work is incomplete, keep it in a draft branchHistorical Trends
Trend: Success rate dipped ~4pp since early April. The increase in WIP and experimental workflow PRs (52 WIP PRs this period vs typical ~30) accounts for most of the decline.
References:
Beta Was this translation helpful? Give feedback.
All reactions