OmniFocus

**Quick judgment** A well-scoped, practically useful skill for users who rely on OmniFocus on macOS. It provides a solid bridge between OpenClaw and OmniFocus via JXA scripts, with good coverage of common GTD workflows and structured JSON I/O that suits agentic use. **Strengths** - Covers the core task lifecycle: add, list (by key filters), search, complete, update, and get stats. - Returns JSON, making it straightforward for agents to parse, summarize, and chain multi-step operations. - Documentation is concrete and example-driven, with recommended interaction patterns (list/search before acting, confirm targets, handle errors clearly). - Aligned with natural language queries users are likely to make about OmniFocus (inbox, flagged, overdue, next actions). **Key risks / limitations** - Hard requirement for macOS with OmniFocus installed and running; unusable in other environments. - Actions like `complete_task` and `update_task` operate by task name, which can be ambiguous; without careful list/confirm steps, an agent could complete or modify the wrong task. - Depends on `osascript` and JXA; performance should be acceptable for personal task lists but is not optimized for very large databases or high-frequency calls. - No built-in dry-run mode or undo; mis-operations affect the user’s real OmniFocus data. **Recommended scenarios** - A user explicitly mentions OmniFocus and wants to capture tasks, review their inbox, see what’s overdue/flagged, or get a daily/weekly summary. - Workflow-style interactions where the agent first lists or searches tasks, shows the user a concise summary, then executes updates/completions only after confirmation. - Productivity coaching or planning sessions where the assistant surfaces OmniFocus statistics (overdue, due-soon, available) and helps the user prioritize next actions. **Use with extra caution when** - Task names are short, generic, or duplicated (e.g., many tasks named “Review” or “Call”), increasing the risk of acting on the wrong item. - The environment may not be macOS with OmniFocus running, in which case the skill will fail and should not be repeatedly retried without checking platform constraints.