[prompt-analysis] Copilot PR Prompt Analysis - 2026-08-27

[github-actions[bot]]

🤖 Copilot PR Prompt Pattern Analysis - 2026-08-27

Summary

Analysis Period: Last 30 days
Total PRs: 1000 | Merged: 830 (83.0%) | Closed: 168 (16.8%) | Open: 2 (0.2%)

Prompt Categories and Success Rates

Category	Total	Merged	Success Rate
Bug Fix	631	497	78.8%
Feature Addition	520	419	80.6%
Testing	385	307	79.7%
Documentation	224	187	83.5%
Refactoring	58	41	70.7%
Other	294	267	90.8%

Note: categories are non-exclusive (a PR can match multiple keyword categories), so totals across categories exceed 1000.

Prompt Analysis

✅ Successful Prompt Patterns

Common characteristics in merged PRs:

• Average prompt length: ~128 words (notably shorter than closed PRs)
• Frequent terms: workflow, run, agent, copilot, comment, actions — task descriptions tend to be tightly scoped to a specific workflow/run/component
• Many merged PRs reference a specific failing job, run ID, or error trace (e.g. "AWF rejects /usr/bin/go...", "MCP gateway aborted during initialization with undefined...")

Example successful prompts:

1. PR Remove symlinked Go bind mount from Avenger #56426: "Avenger's gVisor sandbox aborts before agent startup because AWF rejects /usr/bin/go as a bind-mount source..." → Merged
2. PR Fix MCP gateway startup failure on non-PR events (Daily Go Test Parallelizer) #56425: "The Daily Go Test Parallelizer workflow (schedule-triggered) failed because the containerized MCP gateway aborted during initialization..." → Merged
3. PR Add diagnostics for GitHub release creation failures #56413: "The Release workflow exposed only HTTP 500 when gh release create failed, leaving insufficient data to diagnose..." → Merged

❌ Unsuccessful Prompt Patterns

Common characteristics in closed PRs:

• Average prompt length: ~209 words — ~60% longer than merged PRs
• Frequent terms: issue, cve, runs, report, daily, call — often generated from automated/scheduled reports (e.g. CVE scans, daily digests) rather than a specific, scoped fix
• More likely to originate from templated issue bodies (<issue_title> boilerplate) or open-ended feature proposals without a concrete acceptance criterion

Example unsuccessful prompts:

1. PR [WIP] Fix MCP gateway health timeout races with backend startup timeout #56273: "---- This section details on the original issue you should resolve <issue_title>MCP gateway health timeout races backend startup timeout..." → Closed
2. PR Add browser-local AI workflow repair #56149: "Adds an in-browser prompt experience that uses local AI inference and the WASM compiler to iteratively produce a valid workflow. - Local inference..." → Closed (broad, exploratory feature)
3. PR Fix pkg/console native/wasm API parity #56089: "pkg/console had drift between native and wasm build-tag twins: wasm exposed NewIndeterminateProgressBar()..." → Closed

Key Insights

• Conciseness correlates with success: Merged PR prompts average ~128 words vs ~209 words for closed PRs — a ~63% gap. Long, multi-part prompts appear harder to land cleanly.
• "Other" (uncategorized) prompts have the highest success rate (90.8%), suggesting narrowly-scoped, non-formulaic tasks (not matching generic bug/feature/test keywords) often correspond to precise one-off fixes that are easy to review and merge.
• Refactoring has the lowest success rate (70.7%) among defined categories — broad "improve"/"optimize" requests without a concrete failure to fix are more likely to stall or get closed.
• Automated/report-driven prompts (CVE reports, daily digests, templated <issue_title> boilerplate) skew toward closures — these often lack the specific, actionable detail (run ID, exact error, file reference) that successful prompts include.

Recommendations

Based on today's analysis:

1. DO: Keep prompts short and specific — reference the exact workflow, run ID, job URL, or error message when reporting a failure (as seen in the top merged examples).
2. DO: Scope refactoring/improvement requests to a single, well-defined outcome rather than "improve X" in general terms.
3. AVOID: Long, multi-section prompts assembled from templated issue bodies (e.g., <issue_title> boilerplate) or automated scan reports without added specificity — these correlate with lower merge rates.

Historical Trends

Date	PRs	Merged	Closed	Success Rate	Top Category
2026-08-27 (today)	1000	830	168	83.2%*	bug_fix
2026-07-07	1000	792	198	80.0%	bug_fix
2026-07-06	1000	793	193	80.4%	testing
2026-07-05	1000	805	191	80.8%	bug_fix
2026-07-04	1000	798	186	81.1%	bug_fix
2026-07-03	1000	801	178	81.8%	bug_fix

success rate = merged / (merged + closed)

Trend: Success rate today (83.2%) is modestly above the July baseline (~80-82%), and merged-prompt length remains notably shorter than closed-prompt length across recent snapshots — reinforcing that concise, specific prompts continue to correlate with successful merges. bug_fix remains the most common top category historically.

---

Generated by Copilot PR Prompt Analysis (Run: 33127488023)

Generated by 🔍 Copilot PR Prompt Pattern Analysis · copilot · auto · 22.2 AIC · ⌖ 4.91 AIC · ⊞ 11.1K · ◷

• expires on Aug 28, 2026, 3:53 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 #56725.