[prompt-clustering] Copilot PR Prompt Clustering — 2026-04-26

[github-actions[bot]]

Summary

Analysis Period: Last 30 days · Date: 2026-04-26 · Run: §24965710998

Metric	Value
PRs Analyzed	1000
Merged	775 (78%)
Closed (not merged)	219
Clusters Identified	6
Silhouette Score	0.0284
Highest Merge Rate	Schema, Validation & Safe Outputs (84%)
Lowest Merge Rate	AI Agent Engines & MCP (65%)

Key Finding: The largest cluster (Firewall & Access Control, 233 PRs) reflects ongoing firewall/access-control rule work. The lowest-success cluster is AI Agent Engines & MCP (65% merged), often involving large multi-file engine/version bumps that require iteration.

Full Cluster Analysis Report

Cluster Analysis

Cluster 1: Firewall & Access Control

Attribute	Value
Tasks	233 (23% of total)
Merged	193 (83% success rate)
Avg Files Changed	25.2
Avg Lines Added/Removed	+218 / -168
Top Keywords	addresses, firewall, blocked, rules, details, owner, triggering

Representative PRs:

• #26095 ✅ fix: add imperative verbs to "Super-linter" and "Cross-repo setup guidance" step names
• #24916 ✅ fix: add network.allowed to schema-feature-coverage and ai-moderator Codex workflows
• #25577 ✅ chore: bump CLI versions — Claude Code 2.1.98, Copilot 1.0.22 (unpin), Gemini 0.37.1

---

Cluster 2: Workflow Infrastructure & CI/CD

Attribute	Value
Tasks	199 (20% of total)
Merged	152 (76% success rate)
Avg Files Changed	18.3
Avg Lines Added/Removed	+594 / -221
Top Keywords	issue, step, agent, workflow, cache, run, artifact

Representative PRs:

• #27948 ✅ Add daily skill-optimizer workflow with artifact handoff and 3-improvement issue output
• #26247 ✅ Token optimization: auto-triage-issues workflow
• #27695 ❌ [WIP] Cap Documentation Unbloat at max_turns:30 and require at least 1 safe output

---

Cluster 3: AI Agent Engines & MCP

Attribute	Value
Tasks	161 (16% of total)
Merged	105 (65% success rate)
Avg Files Changed	50.9
Avg Lines Added/Removed	+474 / -267
Top Keywords	mcp, cli, copilot, codex, engine, gateway, claude

Representative PRs:

• #25979 ❌ [WIP] Update CLI version constants for multiple tools
• #27764 ❌ Bump default Claude/Copilot/Codex/GitHub MCP versions and recompile workflow locks
• #28401 ✅ chore: bump Copilot CLI → 1.0.36, Codex CLI → 0.125.0, GitHub MCP Server → v1.0.3

---

Cluster 4: Schema, Validation & Safe Outputs

Attribute	Value
Tasks	154 (15% of total)
Merged	129 (84% success rate)
Avg Files Changed	20.4
Avg Lines Added/Removed	+579 / -325
Top Keywords	import, safe, schema, handler, string, validation, field

Representative PRs:

• #26915 ✅ Add team reviewer support to create-pull-request and add-reviewer safe outputs
• #27193 ✅ Add experimental gated merge-pull-request safe-output with policy-driven merge enforcement
• #27476 ✅ Support extending safe_outputs dependencies via safe-outputs.needs

---

Cluster 5: Testing & Quality Assurance

Attribute	Value
Tasks	142 (14% of total)
Merged	108 (76% success rate)
Avg Files Changed	28.8
Avg Lines Added/Removed	+587 / -218
Top Keywords	test, sha, workflow, awf, tests, retry, job

Representative PRs:

• #27419 ✅ Harden gh aw add SHA pinning, enforce .md/.lock.yml frontmatter hash parity
• #25745 ✅ Fix test failures after action pin updates
• #27645 ✅ Fix Workflow Tools & MCP integration failures after network.firewall deprecation

---

Cluster 6: Repository Management & Docs

Attribute	Value
Tasks	111 (11% of total)
Merged	88 (79% success rate)
Avg Files Changed	12.6
Avg Lines Added/Removed	+372 / -170
Top Keywords	repo, memory, workflow, activation, job, run, docs

Representative PRs:

• #28379 ✅ docs: remove "same-repo only" limit from push-to-pull-request-branch, document cross-repo usage
• #26336 ✅ fix: guard activation .github checkout against cross-repo GITHUB_TOKEN failure on workflow
• #27683 ✅ Add smoke-ci coverage for cache/repo memory and update safe outputs

---

Cluster Comparison Table

#	Theme	PRs	Merge Rate	Avg Files	Avg +Lines	Top Keywords
1	Firewall & Access Control	233	83%	25	+218	addresses, firewall, blocked, rules
2	Workflow Infrastructure & CI/CD	199	76%	18	+594	issue, step, agent, workflow
3	AI Agent Engines & MCP	161	65%	51	+474	mcp, cli, copilot, codex
4	Schema, Validation & Safe Outputs	154	84%	20	+579	import, safe, schema, handler
5	Testing & Quality Assurance	142	76%	29	+587	test, sha, workflow, awf
6	Repository Management & Docs	111	79%	13	+372	repo, memory, workflow, activation

Key Findings

1. Firewall & access control is the single largest category (233 PRs, 83% merged), showing sustained investment in network policy management.
2. AI Agent Engines & MCP has the lowest merge rate (65%), driven by frequent version-bump PRs that often need multiple attempts or are superseded.
3. Schema, Validation & Safe Outputs tasks are highly successful (84% merged) and tend to be well-scoped, contract-driven work.
4. Testing & QA PRs touch the most files on average (28.8 avg), reflecting broad regression coverage additions.
5. The overall 78% merge rate across 1,000 copilot PRs indicates the agent pipeline is producing high-quality, accepted work at scale.

Recommendations

1. Version bump automation: The low merge rate in the AI Agent Engines & MCP cluster suggests version-bump PRs should be batched and have auto-retry logic to reduce churn.
2. Firewall task templates: Given the volume (23% of all PRs), standardised firewall-rule PR templates would speed review and reduce error rates.
3. Test scoping: Large file counts in testing PRs indicate broad but potentially unfocused coverage; consider scoping test PRs to specific subsystems.
4. Schema-first development: The high success rate of schema/validation PRs confirms the value of contract-driven design — expand this pattern to other clusters.

References:

• §24965710998

Generated by Copilot Agent Prompt Clustering Analysis · ● 249.2K · ◷

• expires on Apr 27, 2026, 8:28 PM UTC

[github-actions[bot]]

This discussion has been marked as outdated by Copilot Agent Prompt Clustering Analysis.

A newer discussion is available at Discussion #28795.