[copilot-agent-analysis] Daily Copilot Agent Analysis - 2026-05-21

[github-actions[bot]]

🤖 Copilot Agent PR Analysis - 2026-05-21

Analysis Period: Last 24 hours
Total PRs (agent_prs_total): 65 | Merged (agent_prs_merged): 42 (72.4%) | Avg Duration: ~116 min

Performance Metrics

Date	PRs	Merged	Success Rate	Avg Duration	Open
2026-05-21	65	42	72.4%	116 min	7
2026-05-20	66	51	77.3%	100 min	0
2026-05-19	68	57	85.1%	157 min	1

Trend: 📉 Success rate declining over the last 3 days (85.1% → 77.3% → 72.4%, a ~15% drop from 2 days ago). Closure-without-merge rate rose to 24.6% today.

Agent Task Texts (65 PRs)

PR #	Status	Task Text (first 100 chars)
#33841	OPEN	Thanks for asking me to work on this. I will get started on it and keep this PR's description up to
#33840	OPEN	- [x] Investigate current protected-files fallback behavior and related tests - [x] Inspect recent w
#33839	OPEN	Thanks for asking me to work on this. I will get started on it and keep this PR's description up to
#33836	MERGED	PR comments from ready_for_review-triggered workflows (Test Quality Sentinel, Design Decision Gate
#33831	OPEN	Three production call sites compared Context.Err() to context.Canceled / `context.DeadlineExceed
#33830	OPEN	7+ workflows each independently implement the same ~50-line cache-memory trending pattern (backfill/
#33829	OPEN	ErrNpmNotAvailable and isErrNpmNotAvailable were duplicated in build-tag-split files even though
#33827	MERGED	The PR Code Quality Reviewer averaged ~797K tokens/run with high variance (449K–1.2M), driven by red
#33826	MERGED	Community-labeled issues (opened by external contributors) had no scoring advantage over internal wo
#33819	MERGED	Both list_workflows and list_label GitHub MCP tools silently ignore per_page: list_label har
#33818	OPEN	~64 bare fmt.Println/fmt.Printf calls in pkg/cli/ made it impossible to statically distinguish
#33817	CLOSED	Thanks for asking me to work on this. I will get started on it and keep this PR's description up to
#33815	CLOSED	The compiler processes workflow files sequentially; with large repos this is a bottleneck. This adds
#33814	MERGED	On some CI runners, docker info exits 0 but docker pull immediately returns `"Cannot connect to
#33813	MERGED	Removes the lint-gate job from the CGO workflow and all needs: [lint-gate] references in downstr
#33810	MERGED	Pressing Ctrl+C while gh aw compile was running had no effect — the process kept compiling all rem
#33808	MERGED	Drops the deprecated infer frontmatter field entirely. It was superseded by `disable-model-invocat
#33807	CLOSED	Smoke CI and Design Decision Gate were emitting many gen_ai spans with no terminal usage rollu
#33805	MERGED	The daily spec review identified a P0 gap in forecast confidence handling: the implementation treate
#33797	CLOSED	The daily breaking-change report identified a high-severity schema break: removal of `on.pull_reques
#33781	CLOSED	Thanks for asking me to work on this. I will get started on it and keep this PR's description up to
#33780	CLOSED	infer is marked deprecated in the workflow schema, but included-file frontmatter accepted it witho
#33779	MERGED	Matt Pocock Skills Reviewer intermittently failed before agent startup when Pre-fetch PR diff pr
#33776	MERGED	Strict-mode validation was surfacing ecosystem-domain guidance as a warning, including the `strict m
#33775	MERGED	Runtime installs were always unconstrained by package publish age, increasing exposure to freshly co
#33770	MERGED	Removes the experimental on.pull_request_reviewer synthetic trigger entirely — no legacy compatibi
#33758	CLOSED	pr-sous-chef was attempting formatting for eligible PRs unconditionally. This change narrows that
#33756	MERGED	process.env.GH_AW_EFFECTIVE_TOKENS returns string / undefined, but parseInt only accepts `stri
#33755	CLOSED	The Daily Semgrep Scan workflow was the only scheduled workflow outright failing in the observed win
#33754	CLOSED	Go Module Review (#go-fan) identified style drift and two
#33753	MERGED	Implements the sub_agent_strategy A/B experiment campaign for agent-persona-explorer, testing wh
#33746	MERGED	checkout overrides were only applied in the agent job, while safe_outputs always checked out wit
#33739	MERGED	The agent job in aw-portfolio-yield was failing during postcompute when agent-produced recommend
#33716	MERGED	Several workflow files were still redefining regexes that already exist in `pkg/workflow/expression_
#33707	MERGED	generate_safe_outputs_tools.cjs accesses toolsMeta.required_field_removals but the property was
#33702	CLOSED	Workflows using on.pull_request_reviewer were running twice — triggered correctly by PR lifecycle
#33699	MERGED	Daily model inventory surfaced a critical 10x overcharge bug and several missing/stale entries in th
#33694	MERGED	All 16 regexp.MustCompile() calls inside function bodies were recompiling regex state machines on
#33687	MERGED	create_pull_request.cjs was a 2441-line monolith with no unit test coverage on its pure utility fu
#33685	CLOSED	LintMonster flagged 38 violations where pkg/ code used log.Printf/log.Print patterns that viol
#33683	CLOSED	The custom linter (make golint-custom) flags functions exceeding 60 body lines. Several large func
#33676	MERGED	Token Usage summary was inflating ΔET/ET because cache_read_tokens were weighted in addition to fu
#33674	MERGED	daily-model-inventory was fetching Copilot model inventory from api-proxy:10000/reflect in a pre
#33673	CLOSED	Thanks for asking me to work on this. I will get started on it and keep this PR's description up to
#33672	CLOSED	Thanks for asking me to work on this. I will get started on it and keep this PR's description up to
#33664	MERGED	This updates gh-aw to the requested MCP Gateway and Firewall releases, including regenerated compile
#33661	MERGED	This PR aligns built-in model alias mappings and Effective Token multiplier data with the latest inv
#33660	MERGED	Package manifests (aw.yml) can now list raw GitHub Actions YAML files (.yml) alongside agentic w
#33659	CLOSED	When protected-files: fallback-to-issue triggers with the default bundle transport (and also `am
#33658	CLOSED	The network egress firewall injects bash: ["*"], which triggered the hasBashWildcardInTools heur
#33657	MERGED	The PR triage workflow had no mechanism for operators to supply custom routing rules without modifyi
#33656	MERGED	The logs tool's --workflow-name filter always failed in the MCP server context because `GITHUB_R
#33655	MERGED	The contribution-check orchestrator carried ~70 lines of static report layout rules + a full example
#33654	MERGED	When create_pull_request safe-output touches .github/workflows/ files, the safe_outputs job cr
#33650	MERGED	create-pull-request docs show max-patch-size and max-patch-files in handler config, but parser
#33647	MERGED	Some agent jobs were being marked failed by Parse MCP Gateway logs for step summary when `rpc-mess
#33646	MERGED	8 workflows flagged as non-compliant already have the required report-formatting guidelines (###+
#33645	MERGED	The generic deprecation walker (warnDeprecatedFrontmatterFields) was already wired into the compil
#33639	MERGED	Without a declarative knob to suppress the closing comment, workflows that post a summary via `add-c
#33629	MERGED	Each MCP tool call result grows the LLM context window by some amount, but the exact cost was invisi
#33628	MERGED	The Token Usage step summary showed a single aggregated row per model, hiding per-turn progression.
#33627	MERGED	The CLI consistency issue was that Cobra Short descriptions with trailing sentence punctuation cou
#33626	MERGED	This updates pkg/cli/mcp_server_add_test.go to improve test quality and coverage for the MCP add
#33625	MERGED	Three high-frequency meta-orchestrators (agent-performance-analyzer, workflow-health-manager, `p
#33623	MERGED	This adds a new gh-aw workflow that validates OTLP telemetry quality across traces, metrics, and log

Notable PRs

Closed without merge (16) ⚠️

• PR #33817: Refactor to extract progressive disclosure guidelines (WIP, closed ~2min after open)
• PR #33781: Refactor workflows to adopt github-guard-policy.md (WIP, closed quickly)
• PR #33673 / #33672: WIP fixes for failing js-typecheck / Lint Gate jobs (both closed unmerged)
• PR #33754: Migrate go-gh to typed clients — closed (likely too large a refactor scope)
• PR #33658: engine.permission-mode as first-class Claude engine config — closed
• 10 additional closure-without-merge PRs (see full list above)

Open PRs ⏳ (7)

All 7 open PRs were created within the last 1 hour — still in active flight, no stalls.

Key Insights

• Success rate is trending downward (85% → 72% over 3 days); the elevated closure-without-merge count (16 today vs. 10–15 prior days) is the main contributor — worth inspecting whether a recent prompt/config change is generating more low-quality drafts.
• Cycle time degraded yesterday but recovered today (157 → 100 → 116 min); duration is not the bottleneck, merge eligibility is.

---

Generated by Copilot Agent Analysis (Run: §26246354721)

Generated by 📊 Copilot Agent PR Analysis · ● 6M · ◷

• expires on May 22, 2026, 6:54 PM UTC

[github-actions[bot]]

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

A newer discussion is available at Discussion #34089.