[prompt-analysis] Copilot PR Prompt Analysis - 2026-06-15

[github-actions[bot]]

🤖 Copilot PR Prompt Pattern Analysis — 2026-06-15

Summary

Analysis Period: Last 30 days (up to 1,000 PRs)
Total PRs: 1,000 | Merged: 792 (79.4%) | Closed: 205 (20.6%) | Open: 3

Overall merge rate is 79.4%, consistent with the 30-day historical average of ~80% across 20 tracked snapshots (2026-05-15 → 2026-06-15).

---

Prompt Categories and Success Rates

Category	Total	Merged	Success Rate
Documentation	22	19	86.4%
Chore / CI	29	25	86.2%
Refactoring	21	18	85.7%
Testing	83	68	81.9%
Bug Fix	624	490	78.9%
Feature Addition	201	158	78.6%
Unclassified	20	14	70.0%

Bug fix is by far the dominant category (62.4% of all PRs), and its 78.9% rate drives the overall average.

---

Prompt Analysis

✅ Successful Prompt Patterns

Common characteristics in merged PRs:

• Average body word count: 182 words
• Optimal length range: 20–49 words (88.9% success) or 100–199 words (80.9%)
• Code references (backtick spans, filenames, symbols): present in 99% of merged PRs

Top action verbs by success rate (title):

Verb	Total	Merged	Rate
optimize	11	11	100%
handle	10	10	100%
move	8	8	100%
correct	5	5	100%
update	28	26	92.9%
remove	52	48	92.3%
replace	22	20	90.9%
resolve	9	8	88.9%
add	178	149	83.7%
fix	175	129	73.7%
extract	9	4	44.4%

Conventional commit prefix rates:

Prefix	Total	Merged	Rate
docs:	38	35	92.1%
fix:	77	66	85.7%
chore:	18	15	83.3%
feat:	39	29	74.4%
(none)	808	634	78.5%

Example successful prompts:

Show examples

1. PR #39437 [bug_fix]: "The Copilot Centralization Optimizer workflow failed during task collection because it authenticated the Copilot Agents API with github.token, which is not accepted by that endpoint. This change switches the auth to the copilot-token instead..." → Merged

2. PR #39266 [bug_fix]: "GitHub Actions caches are branch-scoped, so treating cache-hit != 'true' as a miss produced incorrect behavior for the daily AIC cache flow. This change makes miss detection semantic (actual restore failures)..." → Merged

3. PR #39280 [bug_fix]: "The Daily Ambient Context Optimizer flagged two workflows with significant redundant prompt content: smoke-copilot had a 1,179-char section duplicating the framework-injected <mcp-clis> block..." → Merged

---

❌ Unsuccessful Prompt Patterns

Common characteristics in closed PRs:

• Average body word count: 210 words (closed PRs average 28 words more than merged)
• Closed PRs mention errors/failures at a slightly higher rate (54.1% vs 47.5% for merged)
• awf-labeled PRs had a 0% merge rate (8 PRs, likely automated/probe runs)
• copilot/lint* branches: 58.3% | copilot/deep* branches: 60.8%

Example unsuccessful prompts:

Show examples

1. PR #39282 [feature]: "The Copilot SDK driver had no mid-turn watchdog — when the agent stalled after report_intent, sendAndWait waited silently for up to 24.5 min (derived from the 25-min action timeout) before the action timeout..." → Closed (complex speculative feature)

2. PR #39299 [bug_fix]: "When create-pull-request-review-comment.target is "*", the handler silently accepted items missing pull_request_number and only failed at finalization time..." → Closed (edge-case handling rejected)

3. PR #39164 [feature]: Wrapped a third-party TTS SDK in a workflow_dispatch workflow for on-demand speech synthesis → Closed (out-of-scope feature)

---

Key Insights

• Verb precision predicts success: "optimize", "handle", "move", "correct", "update", "replace", "remove" all yield 88–100% merge rates. "fix" — the second most-used verb — only reaches 73.7%. Choosing a precise, scope-limited verb signals a well-understood change.

• Prompt length sweet spot is 20–49 words: Counterintuitively, short-but-specific prompts (88.9%) outperform long ones (200+ words = 78.8%). Closed PRs average 28 more words than merged ones, suggesting verbosity often accompanies under-specified or over-ambitious scope.

• Structured commit prefixes lift quality: The docs: prefix yields 92.1% and fix: 85.7%, both above the 78.5% baseline for unstructured titles. Conventional commits signal intent clearly and correlate with more disciplined, merge-ready PRs.

---

Recommendations

Based on today's analysis:

1. DO: Use scope-specific verbs — update, remove, replace, optimize, handle — instead of the generic fix when the change type permits.
2. DO: Keep PR body prompts to 20–100 words: be specific about what and why, but avoid verbose preamble.
3. DO: Prefix titles with conventional commit types (docs:, fix:, chore:) — they predict a 5–14 pp lift in merge rate over unstructured titles.
4. AVOID: Exploratory or investigative branches (copilot/deep*, copilot/lint*) — these show 58–61% success vs 82–100% for targeted action branches.
5. AVOID: Very long prompts (200+ words); closed PRs averaged 28 more words than merged ones.

---

Historical Trends

Date	PRs	Merged	Closed	Success Rate	Key Insight
2026-06-14	1,000	791	202	79.1%	update/remove verbs 92%+; WIP tags drop to 34%
2026-06-13	1,000	792	204	80.0%	—
2026-06-12	1,000	786	206	79.2%	—
2026-06-15 (today)	1,000	792	205	79.4%	optimize/handle/move 100%; short prompts 88.9%

30-day average: 80% across 21 daily snapshots. Merge rate is stable; verb and length patterns are persistent signals.

---

Generated by Copilot PR Prompt Analysis · Run §27577956135

Generated by 🔍 Copilot PR Prompt Pattern Analysis · 425.1 AIC · ⌖ 21.1 AIC · ⊞ 20.2K · ◷

• expires on Jun 16, 2026, 1:48 PM UTC-08:00