v2.1.207

2.1.207

(☝️ Click to see full diff.)

+6,150 tokens

• NEW: Data: Structured tool output field schema — Documents the per-tool tool_use_result output contract, including completed Agent/Task reports and run totals, and tells clients to render structured output instead of parsing model-facing result text.
• NEW: Skill: /morning slash command — Adds a concise, styled morning-brief workflow that gathers connected calendar and communication data into a safe self-contained HTML artifact, supports user-selected sections and roles, and can configure or update a weekday recurring /morning task with verified local-time scheduling.
• NEW: System Reminder: ClaudeDesign project grant unavailable without verified identity — Explains that project-wide Claude Design approval cannot be offered when the project identity cannot be verified or rendered safely, and directs agents to read and retry a fresh connection once or fall back to per-batch approvals.
• NEW: Tool Description: ScheduleWakeup delay and reason guidance; Tool Description: Snooze (delay and reason guidance); and Skill: /loop self-pacing mode — Centralize wake-up pacing guidance in ScheduleWakeup, add quiet-tick noop reporting and billing-aware prompt-cache advice, retain specific user-visible reasons, and clarify that fallback heartbeats should follow task cadence rather than assuming a five-minute cache window.
• REMOVED: Tool Description: Bash (sandbox — user permission prompt) — Removes the standalone note that disabling the sandbox prompts the user for permission.
• Skill: Auto mode setup — Expands setup to review and safely remove ignored or destructive permissions.allow rules, migrate user-confirmed legacy auto-mode entries out of repo-writable .claude/settings.local.json, preserve hard_deny and deny settings, incorporate untrusted sibling-repository documentation gathered through gh, and keep project-specific proposals narrowly worded while storing supported auto-mode configuration in user settings.
• System Prompt: Harness instructions — Makes system-reminder tag guidance conditional on the active tool context while retaining the instruction to treat hook output as user feedback.
• Tool Description: Artifact — Allows proactive private publishing of ordinary agent-authored work while requiring complete inspection of externally authored files and forbidding publication of sensitive, impersonating, fabricated-record, deceptive credential/payment, or private-individual content, without suggesting alternate hosting after refusal.