[sebyx07]

Monthly improve checkup

• .dz/maintainer/maintainer.yml: schedules: improve, days: [mon], hours_utc: [8]. The schema has weekdays, not days of the month, so it fires every Monday. The brief acts only on the first Monday (UTC day 1-7) and stops on the others. Validated against maintainer.v1.json.
• .dz/briefs/improve.md: reads the signals first (gate on fresh main, .bun-version vs the newest Bun, since Dependabot ignores bun, bun outdated majors, bun run coverage, open bug issues, expectedRed pins that could be unpinned, dated docs claims). Then it picks at most three improvements, in this order: broken or flaky, Bun and dependency majors, correctness (X_* codes, no swallowed errors), measured speed and bytes (axiom 9), docs drift, simplification. Each one ships as its own small PR with evidence. It never bumps the version. It never ships a breaking change: it files an owner issue instead. This is deliberately narrower than escalate:, because an improvement nobody asked for is not worth a migration for every app. It never touches release or publish automation. If nothing is worth doing, it does nothing.
• .dz/README.md lists the new brief. health.md is unchanged: its rows are pinned to scripts/health.ts by health-doc.test.ts, and the monthly check is not a health signal.

Hidden-error audit

Where	Finding	Action
wiki.yml push step	git clone … 2>/dev/null: every clone failure (auth, network, 5xx) was hidden and reported as "Wiki not initialised"	fixed: git's output is printed, and only a not found answer is called an uninitialised wiki. Anything else is Wiki clone failed
docker/deploy-proof/run.sh deploy()	x deploy … > out || true dropped the exit status. A non-JSON answer killed the script at jq under set -e with no refusal	fixed: requires exit 0 and deployed, and the refusal names the exit status, rollout status and findings. Simulated locally with a stub for the deployed, failed, garbage and empty cases. x deploy exits non-zero only when helm does, so a real green proof is unaffected
run.sh package loop	bun pm pack … >/dev/null 2>&1: a failed pack ended the job with no message	fixed: stderr is kept
registry-audit.yml	continue-on-error: true on the audit	deliberate: the next steps key on outcome, file the alarm, then exit 1. Added a reason comment
release.yml, deploy-social-demo.yml || true on gh api	a transient API failure is read as "no verdict"	deliberate, documented, fails closed
release.yml npm view … 2>/dev/null || true	an absent version leads to publish	deliberate: a re-run publish is idempotent and its errors are handled
release-tag.yml gh api …/ref/tags 2>/dev/null	existence probe for the tag	deliberate: a wrong "absent" fails closed at the ref create, which re-probes
ci.yml grep -c … || true, jq … || true render, merge --json || true	grep exits 1 on a zero count. The others are summary renders	deliberate: the verdict is exit "$status" from the real command
ci.yml fail-fast: false	keeps the other parts running when one goes red	deliberate. ci.yml already has no continue-on-error
TS .catch(() => undefined) / catch {} (packages, scripts)	spot-checked: teardown of streams and servers, JSON reads followed by an explicit shape check or refusal, best-effort browser code (localStorage, SW preload)	deliberate, most with a reason comment. No change
skipIf	skipIf(true) only in tests of the skip-cleanup guard itself. Live suites skip on unset services, but the verify-floor ratchet turns a whole-suite skip into a failure	deliberate

Verification

• bun test over the workflow-shape, health-doc, release-due-doc, deploy-proof-due, bun-pin, gate-steps and scaffold-gate tests: 201 pass, 0 fail
• bun run lint: clean apart from 1 existing warning (scripts/lib/health-doc.test.ts:172, also on main)
• bun run doc-paths: every path resolves
• shellcheck -S warning docker/deploy-proof/run.sh: clean
• The deploy-proof job runs on main only, so the run.sh change gets its first real run after merge

🤖 Generated with Claude Code

[developerz-ai]

✅ Reviewed — nothing blocking

0 actionable comment(s) · grounded on your code

_{⚠️ 1 config note(s): test files were not shown to the reviewer — this diff touched no path recognised as a test, so no assertion was available to read the change against}

_{⏱ 54s wall clock · gpt-6.1-sol via router · 2 model call(s) · 391 output token(s)}

_{🤖 developerz.ai — automated review, running on your model and your box.}

[developerz-ai]

Review summary — 6 file(s), 0 finding(s).

Completed review adds a monthly improvement brief and schedule, exposes wiki clone diagnostics, and preserves deployment exit failures. No concrete defects found in the provided diff.

• test files were not shown to the reviewer — this diff touched no path recognised as a test, so no assertion was available to read the change against

🤖 developerz.ai review — automated, running on your model and your box. What is this? · Reviewed by router/gpt-6.1-sol on box hv-dedicated-dev-marshall — 64.3k in / 391 out over 2 calls; cost unknown: no price on record for router/gpt-6.1-sol.