[prompt-analysis] Copilot PR Prompt Analysis - 2026-07-21

[github-actions[bot]]

🤖 Copilot PR Prompt Pattern Analysis - 2026-07-21

Summary

Analysis Period: Last 30 days (2026-06-30 → 2026-07-21)
Total PRs: 1000 | Merged: 764 (76.4%) | Closed: 230 (23.0%) | Open: 6 (0.6%)

Prompt Categories and Success Rates

Category	Total	Merged	Closed	Success Rate
Bug Fix	593	460	127	78.4%
Refactoring	105	84	21	80.0%
Feature Addition	39	31	8	79.5%
Testing	82	61	21	74.4%
Documentation	152	108	44	71.1%
Other	29	20	9	69.0%

✅ Successful Prompt Patterns

Common characteristics in merged PRs:

• Average prompt length: ~188 words
• Most common title keywords: feat, workflow, docs, refactor, copilot
• Top action verbs: fix (221), add (140), refactor (50), remove (30), test (29)

Refactoring prompts lead with 80% success rate — prompts that describe a specific problem with exact metrics and files to change perform best.

Example successful prompts

PR #47061 (Merged): refactor(workflow,cli): eliminate 680 function-length lint findings via targeted helper extraction — References make golint-custom output with exact finding count (680), file paths (pkg/workflow, pkg/cli), and worst-offender ranges (200–96 lines).

PR #47056 (Merged): Improve signed-commits push refusal rendering in fallback issue bodies — Describes exact failure condition and specific output location.

PR #47055 (Merged): fix(eslint): patch switch fall-through false negative and hoisted function false positive in require-return-after-core-setfailed — Names two specific precision gaps, the rule involved, and the exact symptom.

❌ Unsuccessful Prompt Patterns

Common characteristics in closed PRs:

• Average prompt length: ~183 words (slightly shorter)
• Top title keywords: docs, workflow, feat, copilot, bump
• Issues: vague scope, engine/dependency switches without clear rationale, bundled concerns

Example unsuccessful prompts

PR #47044 (Closed): feat(q): apply shared quality gate to Q workflow — References another PR number as context but doesn't describe the expected outcome.

PR #46986 (Closed): Add Copilot AI-credits fallback pricing and skip AI-credits guardrails for BYOK — Two distinct concerns bundled into one prompt.

PR #46928 (Closed): Switch AI Moderator engine from codex to copilot — Engine migration without specifying config files, expected behavior changes, or test criteria.

Key Insights

• Specificity drives success: Merged PRs most commonly include exact counts, file names, rule names, or error messages. Vague scope correlates with closure.
• Refactoring beats bug fixes in success rate (80% vs 78.4%): Well-scoped refactor prompts with clear before/after metrics outperform even targeted bug fixes.
• Documentation has the lowest success rate (71.1%): Doc PRs often lack a clear definition of done.
• Bundled concerns hurt: Closed PRs frequently combine two distinct changes. Single-concern prompts merge more reliably.

Recommendations

1. DO include specific metrics or counts (e.g., "680 lint findings", "3/5 recent runs failed").
2. DO name the exact files, rules, or functions affected.
3. DO describe the specific symptom, not just the fix.
4. AVOID bundling two unrelated changes in one prompt — split into separate PRs.
5. AVOID referencing other PR numbers as primary context without summarizing what changed.

Historical Trends

Date	PRs	Merged	Success Rate
2026-07-21	1000	764	76.4%
2026-07-20	1000	755	75.5%
2026-07-19	1000	762	76.2%
2026-07-18	1000	757	75.7%

Trend: Success rate has been stable in the 75–77% range. Refactoring and feature PRs remain most reliable. Documentation consistently underperforms.

---

References: §29867942549

Generated by 🔍 Copilot PR Prompt Pattern Analysis · sonnet46 29.2 AIC · ⌖ 13.1 AIC · ⊞ 7.6K · ◷

• expires on Jul 22, 2026, 1:03 PM UTC-08:00