Open Source LLM TODO Skill A Fu Uses Claude Code and Codex to Automate Knowledge to Scheduling

A Fu, an open source LLM TODO Skill, turns loose inbox items into structured Markdown task cards through direct integration with Claude Code and Codex.

The tool reads incoming material, detects missing details such as video links, and fills gaps by calling yt-dlp and a local Whisper instance for subtitle extraction. Once the card is complete it supports bulk scheduling, AI-driven grouping, drag-and-drop adjustments in a weekly view, and one-click sync to either the Mac calendar or Feishu calendar.

Users install the package with a single command. The project is hosted on GitHub under the Fable5 API framework.

The workflow begins in the inbox. A Fu parses each entry and writes a Markdown file that already contains due dates, tags, and required context when available. When a video link appears without a transcript, the system queues the download and transcription steps automatically. The resulting subtitle text is appended to the card before the task is shown in the weekly planner.

Batch scheduling lets teams select multiple cards and assign them across days with a single instruction. An AI merge step then combines duplicate or related items into one card, reducing noise in the calendar view. The weekly board supports mouse or trackpad reordering so priorities can shift without editing text files by hand.

Sync options cover both personal and enterprise calendars. The Mac calendar receives events with time blocks preserved. Feishu users receive the same events inside the corporate workspace, keeping personal and work schedules aligned in one place.

The author built A Fu on top of the API version of Fable5 and Codex to keep all logic inside local scripts. No cloud service stores the task data. Calendar tokens are handled through standard OAuth flows and never leave the device except for the chosen sync target.

Developers who want the same pattern can fork the repository and extend the extraction rules for other media types. The one-command install pulls the required Python packages, sets up the Whisper model, and registers the skill with the local Codex instance.

The project addresses a common gap between note-taking and actual execution. Many knowledge workers collect articles, videos, and links yet lack an automated path from collection to scheduled action. A Fu places that path inside an open repository so others can audit or adapt the steps.

Future updates listed in the repository aim to add support for additional subtitle languages and deeper integration with local large language models beyond Claude Code. The current release already demonstrates how three components, an inbox parser, a media extractor, and a scheduler, can sit inside one Skill file.

Developers testing the install report that the weekly view updates within seconds after a sync command. Cards that contain incomplete fields are highlighted so users can decide whether to let the automation continue or to supply the missing data manually.

No pricing is attached to the repository. The entire stack runs on the user’s hardware once the one-line install finishes. Calendar services remain the only external dependency.

The approach shows that targeted skills built on existing LLM runtimes can close the loop between information intake and time allocation without requiring new proprietary platforms.