[prompt-analysis] Copilot PR Prompt Analysis - 2026-04-30

[github-actions[bot]]

Summary

Analysis Period: Last 30 days
Total PRs: 1,000 | Merged: 778 (77.8%) | Closed: 218 (21.8%) | Open: 4 (0.4%)

Overall merge rate is 77.8% — essentially flat vs prior days (78.1% on Apr 27, 78.1% on Apr 26), continuing a gradual downward trend from 80.2% on Apr 17.

---

Prompt Categories and Success Rates

Category	Total	Merged	Success Rate
Bug Fix	884	692	78.3%
Feature Addition	105	78	74.3%
Refactoring	8	6	75.0%
Testing	2	1	50.0%
Other	1	1	100.0%

Note: "Bug Fix" is broadly defined — any prompt with "fix", "resolve", "error", etc. — making it the dominant category at 88% of all PRs.

---

✅ Successful Prompt Patterns

Common characteristics in merged PRs (778 PRs):

• Average prompt length: ~1,196 words
• Prompts tend to include detailed context — error messages, reproduction steps, or references to specific files
• Structured prompts with clear problem descriptions and expected outcomes perform well

Example merged PRs:

1. PR fix(daily-community-attribution): cap Tier 3 lookups and add token budget guardrails to stop runaway #29383: "The Daily Community Attribution Updater hit 159 turns and 8.7M tokens on 2026-04-30 — ~12% of the daily Copilot budget — with a 100% error rate..." → Merged ✅
2. PR Introduce shared/daily-issue-base.md for scheduled issue-creating workflows #29374: "Multiple daily workflows that create GitHub issues duplicate the same boilerplate: activation-app import, reporting import..." → Merged ✅
3. PR fix(smoke-claude): resolve only unresolved review threads in test #16 #29373: "Smoke Claude was failing at safe_outputs when resolve_pull_request_review_thread targeted a thread already resolved by a prior run..." → Merged ✅

---

❌ Unsuccessful Prompt Patterns

Common characteristics in closed PRs (218 PRs):

• Average prompt length: ~1,091 words (~9% shorter than merged)
• Closed PRs often involve large-scale migrations or decompositions that are too broad in scope
• WIP/draft-style prompts ("[WIP]") correlate with closures
• Prompts that attempt to rewrite many files simultaneously tend to fail review

Example closed PRs:

1. PR Migrate raw fmt.Fprintf/Fprintln(os.Stderr) to console.Format* helpers across pkg/ #29358: "139 source files were bypassing the project's console.Format* abstraction by writing directly to os.Stderr..." → Closed ❌ (Too broad: 139 files at once)
2. PR [WIP] Decompose 6 source files exceeding AGENTS.md limit #29357: "[WIP] Decompose 6 source files exceeding AGENTS.md limit..." → Closed ❌ (WIP + multi-file decomposition)
3. PR feat: add repo-mind-light to daily issue/PR agentic workflows #29356: "Integrates the githubnext/repo-mind-light-aw shared workflow..." → Closed ❌

---

Key Insights

• Specificity wins: Prompts that include exact error messages, token counts, or reproduction steps (like PR fix(daily-community-attribution): cap Tier 3 lookups and add token budget guardrails to stop runaway #29383) are significantly more likely to merge. Vague or broad scope prompts struggle.
• Scope creep kills PRs: PRs touching 100+ files (like 139-file migrations) have very low merge rates. Narrower, focused changes succeed better.
• Prompt length correlates positively with success: Merged prompts average 1,196 words vs 1,091 for closed — more context helps the agent make better decisions.
• Bug fixes dominate and perform well: 88% of all prompts are classified as bug fixes, with a 78.3% success rate — aligning well with the overall merge rate.

---

Recommendations

1. DO: Include specific error messages, line numbers, or token/resource metrics in your prompt — Copilot performs better with concrete constraints
2. DO: Keep the scope narrow — target one file or one subsystem per PR; avoid bulk migrations (139 files → split into 10-15 file chunks)
3. DO: Write prompts over ~200 words — more context leads to better outcomes
4. AVOID: WIP prefixes and multi-objective prompts — "[WIP] Decompose 6 files AND fix validation AND update tests" creates ambiguity
5. AVOID: Prompts that require modifying the entire codebase at once — prefer incremental, reviewable changes

---

Historical Trends (Last 7 Analysis Runs)

Date	PRs	Success Rate	Trend
2026-04-17	1,000	80.2%	—
2026-04-23	1,000	78.8%	↓ -1.4pp
2026-04-24	1,000	78.1%	↓ -0.7pp
2026-04-25	1,000	78.2%	→ +0.1pp
2026-04-26	1,000	78.1%	→ -0.1pp
2026-04-27	1,000	78.0%	→ -0.1pp
2026-04-30	1,000	77.8%	↓ -0.2pp

Trend: Merge rate has declined ~2.4pp since Apr 17. The rate has stabilized around 78% for the past week with a slight downward drift. Feature addition success rate (74.3%) dropped compared to Apr 27 (76.2%), while bug fix rate (78.3%) remains stable.

---

References:

• §25190746859

Generated by Copilot PR Prompt Pattern Analysis · ● 448.1K · ◷

• expires on May 1, 2026, 9:47 PM UTC