docs: Claude Code v2.1.272 - workflows pause on usage limits, omitClaudeMd, new financial-advisors repo

[claude-yolo]

Night shift report

WHY THIS MATTERS: Two releases landed (2.1.271 + 2.1.272) with ~70 bug fixes and several new capabilities that change how power users run workflows and subagents. The standout: dynamic workflows now pause when you hit your usage limit instead of failing — a meaningful reliability improvement for long-running jobs. A new Anthropic GitHub repo (claude-for-financial-advisors) surfaced in discovery.json, signaling a new product vertical that isn't archived yet. The API schema for model fallback responses also changed (model field is now nullable), which matters to anyone parsing those responses.

HIGHLIGHTS:

• Workflows pause on usage limits (v2.1.271+): when an agent in a dynamic workflow hits a claude.ai usage limit, the run pauses and auto-resumes when the limit resets — no more failed runs from rate limits. Documented in workflows.md with conditions (interactive sessions, autoContinueAtUsageLimit on, limit resets within 24h, max 2 waits per run). Previously these agents just failed.
• omitClaudeMd for subagents (v2.1.271+): new frontmatter field lets custom/plugin subagents launch without user, project, and local CLAUDE.md files. Useful for tightly scoped subagents that take everything from the delegation prompt. Managed policy files still load. Documented in sub-agents.md.
• Discovery: anthropics/claude-for-financial-advisors appeared in github_org_anthropics.not_fetched — a new public Anthropic repo not yet in sources.json. Suggests a financial-advisors product vertical. A human should decide whether to archive it.
• Dynamic workflow default size change: Pro plan users now default to small (<5 agents) instead of medium; medium now means <10 agents instead of <15. Changes default cost for untuned workflows on Pro.
• API: model field now nullable in fallback request usage.iterations entries — Model → Model or null in the beta API spec. Relevant to anyone parsing fallback usage responses.
• Fast mode in Remote sessions: the host's fast-mode setting or /fast applies in cloud and self-hosted runner sessions where the org allows it.
• Per-command allowed_domains in sandboxed auto mode: Bash/PowerShell/Monitor commands in auto mode with sandboxing can now declare which hosts they need; other hosts are refused for that command.
• multiplier up to 10 in modelPricing: marked-up internal chargeback rates now supported up to 10× in the managed setting and Claude apps gateway pricing block.
• What's new page: weeks 35–37 (Aug 24 – Sep 11) added, covering claude plugin eval, Fable 5.1 availability, computer use in Desktop on macOS, /skill-doctor, /resume for terminal sessions in Desktop, and more.
• Monitoring: event.sequence semantics clarified: it's per-process, not per-session. /clear doesn't reset it; resumed sessions can have non-monotonic values within a session. New event.sequence correlation attribute documented in the event correlation table.
• 70+ bug fixes spanning Bash permission checks, MCP OAuth, Remote Control, Claude Tag, VSCode, cloud sessions, rendering, and more.
• Tombstones: only the date changed — no pages newly gone and none came back.

---

Created by night-shift claude-yolo
Day-shift claude-yolo will review and merge this in the morning