[Weekly Summary] Weekly Issue Summary — 2026-10-05 to 2026-10-12

[github-actions[bot]]

Window: 2026-10-05 15:04 UTC → 2026-10-12 15:04 UTC (rolling 7 days). Charts use 30 days of repository issue history (2026-09-12 → 2026-10-12).

Overview

github/gh-aw opened 1,512 issues in the last 7 days. Almost all of them are automation: 1,457 (96%) from github-actions / Copilot, and 55 from human authors. 1,439 of those new issues are already closed; 73 remain open (3 human, 70 bot — mostly created today).

Metric	Last 7 days	Last 30 days
Issues opened	1,512	5,617
Issues closed (any age)	—	5,804
Still open (of those opened)	73	—
Open backlog (snapshot)	135	peak 407 (Oct 3)
7-day backlog change	−122 (−48%) ⬇️	—
Typical time-to-close (median)	~0.5–1.0 day	~0.53–0.66 day

Human work this week clustered on model routing (@SivaKesava1, 41 issues) and NVX / compiler / firewall (@lpcox, 10). Bot volume is dominated by [aw] workflow-failure tickets (751) plus [WIP], [deep-report], [squad:plan], and cookie/automation chatter.

Three human issues are still open — all model-routing test gaps:

• #67796 — Add a routing smoke for AWF steering notices (AI-credit budget) — @SivaKesava1
• #67795 — Add a Claude-engine model-routing smoke (smoke-claude-routed) — @SivaKesava1
• #67789 — Routing smokes never run the default threat-detection engine — @SivaKesava1

Key metrics

Opened per day (this window)

Date (UTC)	Opened	Bot	Human	Still open
Oct 5 (partial, from 15:04)	53	53	0	0
Oct 6	260	251	9	0
Oct 7	299	295	4	0
Oct 8	207	191	16	1
Oct 9	187	179	8	0
Oct 10	161	154	7	1
Oct 11	209	200	9	12
Oct 12 (through 15:04)	136	134	2	59

Peak creation was Tue Oct 7 (299). Oct 12 looks quieter only because the day is incomplete; many of today’s [aw] failure issues are still open because they have not yet been auto-closed.

Authors (issues opened)

Author	Count	Role
github-actions	1,455	bot
@SivaKesava1	41	human
@lpcox	10	human
@JeffreyCA	2	human
copilot-swe-agent	2	bot
@pelikhan	1	human
@dsyme	1	human

Top labels (issues opened this week)

Label	Count	Notes
agentic-workflows	820	Failure/status tickets from agent runs
automation	349	Scheduled reporters
cascade-suspected	309	Recurring/cascading workflow failures
cookie	198	Short-lived automation cookies
testing	122	Includes human model-routing smokes
code-quality	82	Lint/quality bots
quick-win / task-mining / improvement	67 each	[deep-report] / [AW Top 10]
squad:plan	57	Planning tickets (all closed)
aw-essential	48	[AW Top 10]
security	40	Firewall bumps + compiler hardening
model-routing	28	Almost entirely @SivaKesava1
bug	27	Mix of human product bugs

📈 Issue Activity Trends

Weekly Activity Patterns

Throughput is high and fairly stable: roughly 1,200–1,530 issues opened per calendar week, with closures tracking opens. The open backlog climbed from ~310 (Sep 12) to a peak of 407 on Oct 3, then burned down hard — −80 on Oct 4, −70 on Oct 5, and −129 so far on Oct 12 — to 135 open issues now (−48% vs 7 days ago). Net 7-day change has flipped from slightly positive in late September to clearly negative this week: the repo is closing more than it opens.

Resolution Time Analysis

Median time-to-close stays under a day (typically ~0.53–0.66 days), which matches ephemeral [aw] / cookie tickets that expire or auto-close within hours. The mean is jumpy: Oct 4–5 and Oct 12 show means of ~5d, ~12d, and ~11d because those days closed a tail of old issues (max lifespan 111d on Oct 5 and 168d on Oct 12). The IQR (p25–p75) usually sits between a few hours and ~1–2 days. Read the median, not the mean, as the “how fast do new issues die?” signal.

Notable patterns

• [aw] failure factory (751 issues, 50% of the week). Largest buckets: generic failed (245), other/misc (241), missing terminal safe output (94), report_incomplete (85), missing required tool (24), timeouts (18), no AIC pricing for model (14), unverified daily AI credits (10), engine 429s (9). Recurring names include Daily Go Test Parallelizer, PR Code Quality Reviewer, Issue Monster, skills reviewers, and Daily Reliability Review.
• cascade-suspected (309) is the second-loudest failure signal after the agentic-workflows label. Same workflows are paging repeatedly rather than failing once.
• Cookie + WIP noise. 198 cookie labels and 172 [WIP] titles (all WIP already closed). These inflate opened/closed counts without changing the real backlog.
• Planning burst. 57 [squad:plan] issues (Mycelium Mail and similar) opened and already closed — a one-shot planning wave, not a persistent queue.
• Human theme is model routing. 28 model-routing labels; 41 issues from @SivaKesava1 covering Copilot/Claude/Codex/pi routing, audit attribution, threat-detection on routed runs, and smokes. Compiler/NVX/firewall work from @lpcox is the other human thread (guest libc, RUNNER_* assets, Cloud Hypervisor, firewall 0.28.49–0.28.51).
• Oct 5 is a partial day in the 7-day window (only issues after 15:04 UTC). Full-day Oct 5 created 207 issues.

Human issues opened this week

55 issues from 5 authors; 52 closed, 3 open. Full list:

All 55 human-authored issues

• #67892 (closed) NVX must select Copilot engine assets compatible with guest libc — @lpcox · bug, copilot, compiler
• #67824 (closed) NVX guest command requires unavailable RUNNER_TOOL_CACHE and RUNNER_TEMP assets — @lpcox · bug, compiler
• #67796 (open) Add a routing smoke for AWF steering notices (AI-credit budget) — @SivaKesava1 · bug, testing, copilot, workflows, model-routing
• #67795 (open) Add a Claude-engine model-routing smoke (smoke-claude-routed) — @SivaKesava1 · enhancement, testing, workflows, model-routing
• #67789 (open) Routing smokes never run the default threat-detection engine, and nothing checks that detection stays unrouted — @SivaKesava1 · bug, testing, threat-detection, model-routing
• #67788 (closed) Add a CI test for the Copilot startup model/endpoint mismatch fail-fast (Select Copilot wire APIs from AWF metadata and surface model mismatches #67487) — @SivaKesava1 · enhancement, testing, copilot, model-routing
• #67756 (closed) Bump gh-aw-firewall to v0.28.51 — @lpcox · security, dependencies
• #67754 (closed) Model routing guide: CLI vs Copilot SDK mode for mixed model families, what allowed-models means, and tested examples — @SivaKesava1 · documentation, copilot, model-routing
• #67752 (closed) Support compiling a unified Cloud Hypervisor primary + static script enclave — @lpcox · bug, enhancement, mcp, compiler
• #67596 (closed) Make AWF budget and timeout steering work in gh-aw runs, and show delivered notices in audit — @SivaKesava1 · bug, cli, compiler, model-routing
• #67594 (closed) Routing smokes fail on main: assertion helper filters on purpose values AWF never writes — @SivaKesava1 · bug, testing, workflows, model-routing
• #67522 (closed) Add model-routing smoke workflows with deterministic checks (routed Copilot CLI, routed pi, routed Copilot SDK) — @SivaKesava1 · enhancement, testing, workflows, model-routing
• #67507 (closed) Copilot SDK mode: routed model passed unqualified, declared sub-agents never loaded, and one wire API shared by every Copilot model — @SivaKesava1 · bug, testing, copilot, model-routing
• #67505 (closed) Safe outputs MCP server fails to start on main: setup.sh does not copy memory_schema_contract.cjs (regression from Add declarative JSON schema validation for memory #67439) — @SivaKesava1 · bug
• #67486 (closed) agentic-workflows skill: add a checklist for model and engine misconfiguration — @SivaKesava1 · model-routing
• #67485 (closed) Copilot harness: choose the wire API from AWF endpoint data, fail fast on a model/endpoint mismatch, and report mismatch errors clearly — @SivaKesava1 · model-routing
• #67479 (closed) Agentic maintenance for the model-routing golden test suite — @SivaKesava1 · enhancement, testing, cli, model-routing
• #67390 (closed) Make repo-memory validation recoverable for invalid baseline state — @dsyme · bug, safe-outputs
• #67279 (closed) Model routing attribution: agent-writable awf-routing-outcome.json overrides runner-written routing in the unified session and audit — @SivaKesava1 · bug, security, cli, model-routing
• #67252 (closed) Tests: compiled routed workflows keep the threat-detection job — @SivaKesava1 · testing, safe-outputs, compiler, threat-detection, model-routing
• #67251 (closed) Tests: cross-module model-routing contract tests and measurable harness routing coverage — @SivaKesava1 · enhancement, testing, model-routing
• #67250 (closed) Tests: golden routed-run fixtures for gh aw audit model routing and token usage — @SivaKesava1 · testing, cli, model-routing
• #67237 (closed) OpenTelemetry: export a sparse set of model-routing attributes and document them — @SivaKesava1 · documentation, enhancement, model-routing
• #67233 (closed) Configure /dev/kvm permissions for declarative NVX like Cloud Hypervisor — @lpcox · bug, workflows, compiler
• #67113 (closed) audit/logs: agent job's final aw_info.json and harness routing outcome are lost (deleted on download, absent from aw_session.jsonl) — @SivaKesava1 · bug, cli, model-routing
• #67079 (closed) audit: model_routing.classifier_cost is 0 and per-agent credit reconciliation warns on routed runs (regression from Report sub-agent failures, resolved models, and per-agent spend #66961) — @SivaKesava1 · bug, cli, model-routing
• #67029 (closed) Bump gh-aw-firewall to v0.28.50 — @lpcox · security, dependencies
• #67021 (closed) Generated setup-uv step: disable Actions caching in agentic jobs — @SivaKesava1 · bug, security, compiler
• #67018 (closed) Cloud Hypervisor: HOME tool state under /workspace/.awf-home can be staged into commits and agent patches — @SivaKesava1 · bug, compiler
• #67016 (closed) AWF --exclude-env misses secret references that aren't written exactly as '${{ secrets.' — @SivaKesava1 · bug, security, compiler
• #67012 (closed) Model routing: Claude-engine runs recorded as 'routing rejected (unsupported_endpoint)' although AWF routing succeeded — @SivaKesava1 · bug, model-routing
• #66989 (closed) Add fail-closed declarative NVX runtime support — @lpcox · enhancement, security, compiler
• #66962 (closed) Model routing: aw_info.json, footers, audit/logs and OTEL report the placeholder model (auto/agent) instead of the AWF-routed model and effort — @SivaKesava1 · bug, safe-outputs, cli, model-routing
• #66960 (closed) audit: sub-agent rows hide failures, ignore model aliases and dated ids, and have no per-agent spend (follow-up to audit: sub-agent attribution is guessed from agent-stdio.log, so real sub-agents and their spend are invisible #66740) — @SivaKesava1 · bug, copilot, cli, model-routing
• #66923 (closed) Bump gh-aw-firewall to v0.28.49 and gh-aw-mcpg to v0.4.30 — @lpcox · mcp, dependencies
• #66788 (closed) Claude on Copilot fails AWF pre-flight: provider gateway URL ignores network isolation — @SivaKesava1 · ai-inspected
• #66787 (closed) Model routing on claude/codex/pi fails closed: AWF selection endpoint and effort are chosen for Copilot CLI, not the engine — @SivaKesava1 · ai-inspected, model-routing
• #66779 (closed) runtimes: uv: setup-uv's UV_CACHE_DIR is unwritable inside the AWF sandbox, and agent jobs save a repo-wide setup-uv cache — @SivaKesava1 · ai-inspected, awf
• #66740 (closed) audit: sub-agent attribution is guessed from agent-stdio.log, so real sub-agents and their spend are invisible — @SivaKesava1 · community, ai-inspected, model-routing
• #66716 (closed) Safe outputs: blank optional fields (e.g. create_pull_request stack_position: "") get MCP "success" but are rejected by the collector, dropping the output — @SivaKesava1 · community, ai-inspected
• #66715 (closed) pi engine silently ignores inline sub-agents and their models; agent fakes delegation on the main model — @SivaKesava1 · community, ai-inspected, model-routing
• #66714 (closed) Model routing: support the Claude, Codex and pi engines on the Copilot API — @SivaKesava1 · ai-inspected, model-routing
• #66639 (closed) Support runtime: cloud-hypervisor for enclaves — @lpcox · enhancement, workflows, compiler
• #66638 (closed) Model routing: show AWF routing decisions in gh aw logs and gh aw audit — @SivaKesava1 · community, ai-inspected, model-routing
• #66637 (closed) gh aw audit, view, outcomes and logs build OWNER\REPO on Windows, so artifact downloads fail — @SivaKesava1 · community, ai-inspected
• #66337 (closed) Docs: add a model routing guide for engine.model-routing — @SivaKesava1 · ai-inspected
• #66293 (closed) Daily AIC guardrail counts completed v0.91.2 runs as zero — @JeffreyCA · community
• #66280 (closed) Copilot sub-agents fail with 400 when their model needs a different API than the main model (Claude under GPT, GPT under Claude) — @SivaKesava1 · ai-inspected
• #66245 (closed) Bump gh-aw-mcpg to v0.4.29 — @lpcox · mcp, dependencies
• #66231 (closed) Model routing: let sub-agents use their own models by emitting routing.candidateModels — @SivaKesava1 · ai-inspected, model-routing
• #66226 (closed) Model routing: routed workflows run stale AWF and router images regardless of the AWF version — @SivaKesava1 · ai-inspected
• #66224 (closed) Model routing: threat detection fails in routed workflows, so safe outputs are applied without a threat verdict — @SivaKesava1 · community, ai-inspected
• #66221 (closed) Model routing: the routing-conversation step reads a prompt path that doesn't exist in the agent job — @SivaKesava1 · ai-inspected
• #65963 (closed) Daily AIC guardrail reports fallback estimates for cancelled runs as consumed usage — @JeffreyCA · community
• #65960 (closed) gh aw add: Markdown security scanner rejects JavaScript package resources containing HTML templates — @pelikhan · bug, security, cli

Bot / automation issues (grouped)

1,457 bot-opened issues this week. Titles are grouped rather than listed one-by-one.

Bot title prefixes (count)

Prefix	Count
[aw]	751
[WIP]	172
(no [...] prefix)	122
[deep-report]	67
[squad:plan]	57
[AW Top 10]	48
[PR Triage Report]	25
[duplicate-code]	12
[lint-monster]	12
[pr-sous-chef]	9
Work <hash>	9
Daily reporters ([model-inventory], [team-status], [aw-compat], [work-queues], [cli-tools-test], [github-docs-seo], [windows-integration], [copilot-cli-research], [Contribution Check Report], [smoke-temporary-id], [uk-ai-resilience])	7 each
[container-image-scan], [pr-review], [ca]	6 each
[agentic-token-optimizer], [cli-consistency], [agentic-token-audit], [go-fan], [model-resolution], [schema-audit]	5 each

[aw] failure breakdown (751)

Failure shape	Count
Generic failed	245
Other / mixed messages	241
Finished without a terminal safe output	94
Reported incomplete result	85
Missing required tool	24
Timed out	18
No AI credits pricing for model	14
Could not verify daily AI credits	10
Engine rate limit (HTTP 429)	9
Missing required data	8
Stale lock file	3

Repeated workflows in the “other” bucket include Daily Reliability Review (429), Issue Monster (429), Impeccable / Matt Pocock Skills Reviewer (timeout), missing-tool reporters, and models with no AIC price (gpt-6.1-sol).

Next actions

1. Triage the 94 missing-terminal-safe-output and 85 report_incomplete [aw] tickets — they are the highest-volume actionable failure modes (not just “failed”). Prefer fixing emitters (noop / terminal safe-output) over filing more issues; consider report-failure-as-issue: false on noisy workflows that already expire next day.
2. Collapse cascade-suspected repeats (309) so one failing workflow does not open a new issue every run. Dedup by workflow_id before insert.
3. Close or land the 3 remaining human issues (#67789, #67795, #67796) to finish the model-routing smoke coverage that dominated this week’s human queue.
4. Keep watching mean vs median TTC. Median < 1 day is healthy; mean spikes on Oct 4–5 and Oct 12 mean old backlog is being swept, which is good — do not treat those days as a regression in new-issue handling.

References: §38267134323

Generated by 📅 Weekly Issue Summary · pi · gro46 · 74.2 AIC · ⌖ 16 AIC · ⊞ 10K · ◷

• expires on Oct 13, 2026, 7:21 AM UTC-08:00