[prompt-clustering] Daily Prompt Clustering Analysis — 2026-05-04

[github-actions[bot]]

Summary

Analysis Period: 2026-04-15 to 2026-05-04 (last ~20 days)
Total PRs Analyzed: 1,000
Clusters Identified: 8
Overall Merge Rate: 77.5% (775/1,000)
Workflow Run: §25313820803

The 1,000 most recent Copilot-created PRs were clustered using TF-IDF vectorization on PR titles and bodies, with k=8 selected via silhouette scoring. The repository shows consistent high-velocity agentic work — averaging 50 PRs/day — spanning eight distinct task themes dominated by general GitHub/workflow improvements (33%), documentation/testing tasks (24%), and MCP-related refactors (11%).

### Cluster Analysis — 8 Task Themes

All clusters had healthy merge rates (68–82%). Notable patterns below.

Cluster 1 — GitHub Workflow & Agentics Maintenance (326 PRs, 75% merged)

The largest cluster covers broad workflow-level improvements: runner config, tag pinning, action version updates, and agentic workflow compilation fixes. Mixed bag of bug fixes and maintenance with solid, if slightly lower, merge rate.

Top terms: workflow, github, agentic, request, pull
Top categories: Bug Fix (34%), Other (33%), Add (14%)
Example PRs: #28150, #28533, #29031

---

Cluster 2 — Tests, Docs, Cache & Copilot Prompts (238 PRs, 82% merged)

Second-largest cluster covers test coverage additions, documentation improvements, cache/memory handling, and Copilot prompt/instruction tuning. High merge rate signals these are well-understood, incremental improvements.

Top terms: test, docs, cache, memory, copilot, perf
Top categories: Bug Fix (43%), Other (33%), Docs (8%)
Example PRs: #28956, #28161, #26507

---

Cluster 3 — MCP Server Refactors & Gateway Fixes (112 PRs, 79% merged)

MCP-focused work: fixing MCP config formats, refactoring validators, resolving gateway auth issues, and separating GitHub-specific logic. The high refactor proportion (37%) indicates active architectural evolution of the MCP layer.

Top terms: mcp, refactor, gateway, github, shared
Top categories: Refactor (37%), Bug Fix (24%), Other (20%)
Example PRs: #27722, #26585, #27102

---

Cluster 4 — CLI & MCP Version Management (74 PRs, 81% merged)

CLI consistency improvements, help-text normalization, version bumps, and MCP CLI bridge enhancements. High merge rate and small-to-medium change sets suggest reliable, well-scoped tasks.

Top terms: cli, mcp, copilot, help, claude, bump
Top categories: Bug Fix (35%), Chore/Deps (18%), Other (16%)
Example PRs: #28842, #26715, #26558

---

Cluster 5 — Feature Development (71 PRs, 76% merged)

Pure feature additions tagged with feat: — new audit commands, schema extensions, experiment infrastructure, and engine capabilities. Slightly lower merge rate than bug-fix clusters, consistent with higher complexity.

Top terms: feat, audit, schema, command, experiments, engine
Top categories: Feature (99%)
Example PRs: #28913, #29783, #26594

---

Cluster 6 — Safe-Outputs System (65 PRs, 68% merged)

The lowest merge rate cluster — safe-outputs validation, noop guidance, manifest alignment, and pull-request-level output constraints. The lower success rate may reflect the complexity of coordinating output semantics across many workflows.

Top terms: safe, outputs, output, pull, request
Top categories: Bug Fix (29%), Other (23%), Add (17%)
Example PRs: #27479, #29270, #29269

---

Cluster 7 — Daily Agentic Workflows (60 PRs, 75% merged)

Daily scheduled workflow management: adding new daily checks, rebalancing engine assignments, refactoring shared daily-* base imports, and recompiling lock files. Steady cadence of infrastructure maintenance.

Top terms: daily, workflow, report, optimizer, workflows
Top categories: Other (33%), Bug Fix (23%), Feature (22%)
Example PRs: #28434, #30001, #29787

---

Cluster 8 — Pre-Agent & Sub-Agent Infrastructure (54 PRs, 81% merged)

Agent lifecycle improvements: pre-agent sanitization, sub-agent orchestration, OTLP span instrumentation, manifest handling, and activation flow fixes. High merge rate indicates tight, targeted changes.

Top terms: agent, pre, feat, steps, sub, engine
Top categories: Other (28%), Bug Fix (26%), Feature (24%)
Example PRs: #29420, #28290, #29668

### Success Rate by Task Category

Categorized by conventional commit prefix / title pattern:

Category	PRs	% of Total	Merge Rate
Bug Fix (fix:)	313	31.3%	81.2%
Other (no prefix)	271	27.1%	81.5%
Feature (feat:)	123	12.3%	79.7%
Add (unprefixed)	95	9.5%	78.9%
Refactor	52	5.2%	78.8%
WIP/Investigation	42	4.2%	9.5% ⚠️
Chore/Deps	34	3.4%	70.6%
Docs	32	3.2%	81.2%
Tests/Coverage	13	1.3%	92.3% 🏆
Update/Maintenance	13	1.3%	76.9%
Security	6	0.6%	83.3%
Close/Resolve	6	0.6%	83.3%

### Daily Activity (PR volume per day)

Date	PRs
2026-04-15	16
2026-04-16	84
2026-04-17	47
2026-04-18	38
2026-04-19	41
2026-04-20	57
2026-04-21	55
2026-04-22	51
2026-04-23	71
2026-04-24	46
2026-04-25	53
2026-04-26	32
2026-04-27	30
2026-04-28	55
2026-04-29	58
2026-04-30	58
2026-05-01	78
2026-05-02	46
2026-05-03	63
2026-05-04	21 (partial)

Average: ~50 PRs/day. Peak on 2026-04-16 (84 PRs), with weekday patterns visible (dips on 2026-04-26/27).

### Recent PRs Sample (last 30)

PR #	Title	Category	Outcome	Δ lines
#30126	feat: auto-allow playwright-cli bash command when playwright cli mode enabled	Feature	🔄 open	+321/-12
#30122	Add mattpocock-skills-reviewer agentic workflow	Other	🔄 open	+1959/-0
#30110	fix: resolve 3 claude-engine workflow failures (safe-output misses + blocked commands)	Bug Fix	🔄 open	+99/-72
#30109	Fix missing safe-output calls in Schema Consistency Checker and Multi-Device Docs Tester	Bug Fix	🔄 open	+71/-21
#30100	Fix stale $INSTRUCTION assertion in TestEngineArgsIntegrationCodex	Bug Fix	🔄 open	+37/-7
#30072	chore: reduce per-engine boilerplate in domains.go public API	Chore/Deps	🔄 open	+78/-132
#30071	refactor: decouple safe-outputs checkout from event trigger context	Refactor	🔄 open	+1293/-78
#30070	fix: propagate context in action SHA resolution to enable timeout/cancellation	Bug Fix	🔄 open	+315/-185
#30060	Add GitHub Copilot billing multipliers collection to daily-model-inventory	Other	🔄 open	+275/-20
#30059	feat: migrate sergo workflow from cache-memory to repo-memory	Feature	✅ merged	+163/-122
#30057	feat: add daily-geo-optimizer agentic workflow for GEO auditing	Feature	✅ merged	+1702/-0
#30054	feat(models): add reasoning/gpt-5-nano aliases, fix multipliers — 2026-05-04 sync	Feature	✅ merged	+56/-11
#30053	refactor: eliminate duplicate utilities and trivial alias functions	Refactor	✅ merged	+22/-67
#30052	[WIP] Fix Daily Model Inventory Checker Copilot CLI silent startup crash	WIP	❌ closed	+0/-0
#30046	Analysis: branch storage supports multiple experiments per workflow ID	Other	✅ merged	+56/-68
#30045	fix(pi): use api-proxy Docker service hostname for LLM gateway routing	Bug Fix	✅ merged	+0/-0
#30044	feat: update daily-experiment-report to use experiments CLI commands	Feature	✅ merged	+150/-119
#30040	fix: compiler detects and sanitizes single-quoted bash tool commands	Bug Fix	✅ merged	+197/-51
#30035	feat: add default codex_harness.cjs with retry logic for Codex engine	Feature	✅ merged	+666/-62
#30032	feat: add api-proxy test coverage for Pi engine	Feature	✅ merged	+222/-89
#30030	fix: single-quote GH_AW_OTLP_ENDPOINTS to prevent YAML sequence parsing	Bug Fix	✅ merged	+132/-33
#30029	feat: extend experiments analyze command with statistical computation	Feature	✅ merged	+1243/-0
#30028	feat: query /reflect before and after running the agent in harnesses	Feature	✅ merged	+95/-18
#30027	Mark experiments as experimental with compiler warning	Other	✅ merged	+133/-0
#30026	fix: resolve TypeScript typecheck errors in JS files	Bug Fix	✅ merged	+6/-9
#30025	Remove owner field from experiments	Other	✅ merged	+2/-22
#30024	docs: W3C-style A/B experiments specification	Docs	✅ merged	+984/-0
#30023	fix: use proper experiments.NAME == "value" syntax in experiment docs	Bug Fix	✅ merged	+9/-4
#30021	feat: add support for multiple OTLP endpoints via polymorphic endpoint field	Feature	✅ merged	+1386/-288
#30020	feat: add hidden experiments command to read experiment state	Feature	✅ merged	+1045/-1

Key Findings

1. High throughput, strong success rate: 1,000 PRs in ~20 days at 77.5% overall merge rate, with most clusters achieving 75–82%. The agentic workflow system is operating at scale with high quality.

2. WIP/Investigation tasks are outliers: 42 PRs tagged [WIP] had only a 9.5% merge rate — these serve as exploratory probes, investigation branches, or staging areas that rarely land directly. This is by design but represents ~4% of volume.

3. Tests/Coverage tasks are highest-confidence: The 13 test-focused PRs achieved a 92.3% merge rate — the highest of any category. Well-scoped testing improvements are highly predictable.

4. Safe-outputs cluster has the most friction: Cluster 6 (safe-outputs) has the lowest merge rate (68%) and concentrates bug fixes and output constraint issues. This is the most complex cross-cutting subsystem and may benefit from more targeted prompt engineering.

5. MCP and refactor work is clean and reliable: Clusters 3, 4, and 8 (MCP, CLI, agent infrastructure) have 79–81% merge rates despite high refactor ratios, suggesting well-structured decomposition tasks.

Recommendations

1. Improve WIP task outcomes: WIP/Investigation PRs rarely convert. Consider a policy of converting WIP findings into scoped follow-up tasks rather than leaving them as dead-end PRs (42 closed PRs = ~4% waste).

2. Invest in safe-outputs prompt specificity: The safe-outputs cluster consistently underperforms. Tighter pre-agent context (e.g., pre-fetching current safe-output config, providing explicit constraint checklists) could reduce the fix/retry loop in this area.

3. Standardize conventional commit usage: 27% of PRs use no conventional prefix, making categorization and automation harder. Consistent feat:/fix:/refactor: prefixes would improve routing and reporting accuracy.

4. Leverage the tests/coverage pattern: Test-focused tasks have the highest merge rate. When introducing new features or refactors, pairing them with an explicit test-coverage sub-task appears to produce the most reliable outcomes.

References:

• §25313820803

Generated by Copilot Agent Prompt Clustering Analysis · ● 339.4K · ◷

• expires on May 5, 2026, 10:53 AM UTC