health digest 2026-09-14

[devin-ai-integration]

Weekly read-only health digest for Hmbown/CodeWhale, window 2026-09-07 → 2026-09-14. No PRs opened/closed, nothing pushed, no branches deleted. No previous health digest issue exists to close.

Recommended actions (ranked by impact)

1. Approve the contribution gate for @BX166 (feat(providers): add AICraft OpenAI-compatible provider template #6171). 5 workflow runs (CI, OpenHarmony, Web Frontend, both review bots) are stuck at action_required; the author is not in .github/APPROVED_CONTRIBUTORS. Approve the runs and add all:BX166 or comment /lgtm. @VincentCorleone (fix(weixin-bridge): make the Weixin bridge runnable and simplify its Quick Start #6170) is also unlisted but currently has no blocked runs.
2. Main CI was red for 56 of 103 ci.yml runs on main this week (54%). Latest main runs are green (last failure a286ded6, Version drift job). The dominant failure was a cluster of 8+ tests failing together on the same commits (open_or_init_with_cap_rejects_oversized_workspace ×12 job failures on main, the deepseek-cn/doctor/runtime_api group ×7) — these are broken-commit failures, not flakes. Only 4 (sha, job) pairs actually flip-flopped (true flakes) — see §3.
3. Land or close the two stale Hmbown PRs (feat(cloud-facts): signed, versioned, cached facts channel (slice 1, flag off) #5752 cloud-facts, 15 days no activity; chore(brand): serve the canonical whale on account-entry pages #5861 brand whale, 11 days): both CONFLICTING, both failing Test (windows-latest). Three other PRs are also CONFLICTING/DIRTY (v0.9.14: console-freeze, approval-death, compaction, and session-retention fixes #6161, Professionalize Computer Use and add its official download page #6134, feat(subagent): rate-limit-adaptive launch scheduling #6055) — main churn is outrunning open work.
4. Version drift: npm/deepseek-tui is 0.8.49 vs workspace 0.9.13; web/package.json is 0.1.0 (probably intentional, but the Version drift CI job failed 14 times this week). Dead-code budget is exactly at budget (413/413) — the next #[allow(dead_code)] fails CI.
5. Docs drift: 14 code slash commands undocumented (/advisor /balance /exit /import-claude /lsp /pin /structcopy /system /transcript /verbose /voice /voicecontrol /voicesend /workbar); 7 documented commands not in code (/rc /dryrun /preview_request /plan /yolo /relaunch /relay). Prune 25 stale bot branches (§2) once confirmed dead.

1. Stale PRs (no commit or comment since 2026-09-07)

PR	Author	Age	Last activity	Failing checks	Unresolved threads	Mergeable
#5752	@Hmbown	15d	2026-08-30	Test (windows-latest)	0	CONFLICTING
#5861	@Hmbown	11d	2026-09-03	Test (windows-latest)	2	CONFLICTING

Other open PRs (8, all active in window): #6171 BX166 — gate blocked, 5 action_required runs; #6170 VincentCorleone — green, 8 unresolved threads, unlisted contributor; #6161 Hmbown — CONFLICTING, 8 threads; #6134 Hmbown — CONFLICTING, Lint failing; #6105/#6104 dependabot — green, mergeable; #6096 aboimpinto — green, CLEAN, 0 threads (ready to land); #6055 h3c-hexin — CONFLICTING, Lint + Windows Test + npm smoke failing, 3 threads. buildkite/codewhale-ci appears in the rollup only on #5861 (state None).

Command

for n in 6171 6170 6161 6134 6105 6104 6096 6055 5861 5752; do gh pr view $n --json author,createdAt,mergeable,mergeStateStatus,reviewDecision,statusCheckRollup,commits,comments; done
# reviewThreads via GraphQL: pullRequest(number:$n){reviewThreads(first:100){nodes{isResolved}}}
gh run list --branch <headRef> --limit 5 --json status,conclusion,name

2. Stale branches (no commits since 2026-08-15, no open PR, snapshot/* skipped) — 25

Prefix	Count	Last commit
copilot/* (copilot-swe-agent[bot] / Hmbown)	14	2026-08-03
codex/*	3	2026-07-31 – 08-01
bisect/win-*	3	2026-08-06
contrib/accepted-0.9.4, contrib/pr-5233	2	2026-08-03
cleanup/v095-manageability, web/copy-buttons-hanspacing, copilot/runtime-api-* etc.	3	2026-08-07 – 08-13

Full list + command

git fetch --prune && git for-each-ref --sort=committerdate --format='%(refname:short) %(committerdate:short) %(authorname)' refs/remotes/origin

origin/codex/v094-budget-5001 2026-07-31 Hmbown
origin/codex/v094-bench-turn-scoping 2026-08-01 Hmbown
origin/codex/wave-5055-effort-table 2026-08-01 Hmbown
origin/copilot/v094-make-responses-provider-profiled 2026-08-03 copilot-swe-agent[bot]
origin/copilot/rename-deepseekclient-legacy-internals 2026-08-03 copilot-swe-agent[bot]
origin/copilot/move-workflow-status-top-bar 2026-08-03 copilot-swe-agent[bot]
origin/copilot/v094-extract-responses-dialect-policy 2026-08-03 copilot-swe-agent[bot]
origin/copilot/fix-provider-switching-issue 2026-08-03 copilot-swe-agent[bot]
origin/copilot/fix-sub-agent-fleet-spawn-rejects 2026-08-03 copilot-swe-agent[bot]
origin/copilot/fix-fleet-config-silhouette 2026-08-03 copilot-swe-agent[bot]
origin/copilot/whaleflow-janitor-implementation 2026-08-03 copilot-swe-agent[bot]
origin/copilot/whaleflow-cached-main-overlay 2026-08-03 copilot-swe-agent[bot]
origin/copilot/zero-sandbox-mode-dev 2026-08-03 copilot-swe-agent[bot]
origin/copilot/add-skill-install-update-uninstall-trust-audit 2026-08-03 copilot-swe-agent[bot]
origin/copilot/runtime-api-add-bounded-mcp-config 2026-08-03 Hmbown
origin/copilot/add-memory-inspection-endpoint 2026-08-03 copilot-swe-agent[bot]
origin/copilot/runtime-api-expose-verifier-evidence 2026-08-03 Hmbown
origin/copilot/expose-persistent-goal-loop-state 2026-08-03 Hmbown
origin/contrib/accepted-0.9.4 2026-08-03 Hunter Bown
origin/contrib/pr-5233 2026-08-03 Hunter Bown
origin/bisect/win-after-5077 2026-08-06 CodeWhale Bot
origin/bisect/win-after-5240 2026-08-06 CodeWhale Bot
origin/bisect/win-debug 2026-08-06 CodeWhale Bot
origin/cleanup/v095-manageability 2026-08-07 CodeWhale Bot
origin/web/copy-buttons-hanspacing 2026-08-13 CodeWhale Bot

3. CI flake ranking (2026-09-07 → 09-14)

GitHub Actions ci.yml: 238 runs, 88 failed. main: 103 runs / 56 failed. pull_request: 122 / 27.

Failed jobs: Lint 53 · Test (windows-latest) 46 · Test (macos-latest) 40 · Test (ubuntu-latest) 15 · Version drift 14 · npm wrapper smoke (ubuntu) 4 · Integrations 3 · Documentation 2 · Safety gate 2 · npm wrapper smoke (windows) 1.

Top failing tests (count = failing-job occurrences across all runs):

#	Test	Count	Example
1	tools::subagent::tests::queued_turn_owned_child_parks_without_a_false_start_transition	36	run
2	tools::subagent::tests::turn_end_parking_preserves_a_step_zero_resumable_checkpoint	36	same
3	runtime_api::tests::fleet_status_runtime_api_exposes_state_and_actions	32	run
4–10	deepseek-cn / doctor_endpoint / provider_switch / composed_frame cluster (7 tests, fail together)	28 each	same
11	snapshot::repo::tests::open_or_init_with_cap_rejects_oversized_workspace	28 (12 on main)	run
12	route::export::tests::golden_providers_export_matches_registry	24	run
13	local_ollama::tests::probe_live_local_ollama_catalog_reads_api_tags	10	run
14	core::engine::tests::first_turn_waits_for_explicit_mcp_schema_without_waiting_for_unrelated_server	10	run
15	tui::keybindings::tests::keybindings_md_and_help_catalog_do_not_drift	6	run

Truly flaky ((sha, job) failed then passed on the same commit): 700b6fc0 Test (ubuntu) · b6b3df36 Test (macos) · 0fa9ee91 Test (ubuntu) + npm wrapper smoke (windows). Everything else above failed consistently on its commit → regression/broken-branch, not flake. probe_live_local_ollama_catalog_reads_api_tags (live network probe) is the most likely environment-dependent candidate.

Buildkite codewhale-ci: 78 builds — 63 passed, 15 failed. Failed steps: :rust: workspace tests 12 · :apple: workspace tests 10 · :rust: fmt + clippy 4. No commit both failed and passed (0 flakes). 6 of 15 failures on fix/0913-release-recovery-20260907, 3 on integration/new-prs-20260907. Example: https://buildkite.com/hunter-bown/codewhale-ci/builds/1929

Commands

gh run list --workflow ci.yml --created '>=2026-09-07' --limit 300 --json databaseId,headSha,headBranch,event,conclusion,createdAt,url
gh run view <id> --json jobs
gh run view <id> --log-failed | grep -E 'test .* \.\.\. FAILED'
curl -s -H "Authorization: Bearer $BUILDKITE_READ_TOKEN" "https://api.buildkite.com/v2/organizations/hunter-bown/pipelines/codewhale-ci/builds?created_from=2026-09-07T00:00:00Z&per_page=100"

4. Drift checks (fresh --depth 1 clone of main)

Check	Result
python3 scripts/check-dead-code-budget.py	PASS: 413 attributes, exactly at budget. (exit 0) — zero headroom
Cargo.toml workspace version	0.9.13
CHANGELOG.md top entry	0.9.13 (2026-09-13) ✓
npm/codewhale, npm/runtime-sdk	0.9.13 ✓
npm/deepseek-tui	0.8.49 (lagging)
web/package.json	0.1.0 (not release-tracked)

Docs vs product — sources of truth: 88 CommandInfo.name entries under crates/tui/src/commands/** (registered via groups::all_command_groups()), help view crates/tui/src/tui/views/help.rs; settings taxonomy ConfigCategory::ALL (7 categories: Appearance, ModelsProviders, Work, ToolsMcp, Trust, Motion, Advanced) in crates/tui/src/tui/views/mod.rs; /setup STEP_SPECS (10 steps) in crates/tui/src/tui/setup/mod.rs. Scanned docs/ and web/ (excluding node_modules).

• In code, not in docs (14): /advisor /balance /exit /import-claude /lsp /pin /structcopy /system /transcript /verbose /voice /voicecontrol /voicesend /workbar
• In docs, not in code (7): /rc /dryrun /preview_request /plan /yolo /relaunch /relay
• Settings: category names are generic words (Work, Trust, Motion…) so a text scan cannot prove docs coverage; no doc page enumerating the 7 ConfigCategory entries or the 10 /setup steps was found — worth a dedicated /settings reference page.

5. Security posture

Source	Open
Dependabot alerts (gh api repos/Hmbown/CodeWhale/dependabot/alerts?state=open)	0
CodeQL / code-scanning alerts	unavailable — HTTP 403 "Resource not accessible by integration"; GITHUB_CODEWHALE_SECURITY_PAT is not provisioned

6. Open issues — 92

Label	Count
enhancement	38
(unlabeled)	35
rust	16
tui / cleanup	9 / 9
documentation	6
bug	5
tools / reliability	4 / 4
agent-ready	3
ux / v0.9.12 / responses-api	2 each
dependencies, subagents, question, external-memory, release-blocker, model-lab	1 each

Issues with no non-author human response and no update in >14 days (before 2026-08-31): 0. Caveat: every issue gets an automated codewhale-issue-intake comment; excluding bots (github-actions, *[bot]) still yields 0 because all 92 were updated within 14 days (stale-bot/label churn), so this metric is not currently discriminating.

Command

gh issue list --state open --limit 1000 --json number,title,labels,createdAt,updatedAt,comments,author

[github-actions]

Thanks @devin-ai-integration[bot] for the report.

This issue is staying open for maintainer triage. CodeWhale gets better because people bring us real edge cases from real machines, providers, regions, and workflows.

If you can add a reproduction, logs, version output, screenshots, or the provider/model involved, that makes it much easier for us to verify and harvest the fix. Maintainers may comment /lgtmi to mark recurring issue reporters as approved so this intake note is skipped next time.