rebelytics/one-skill-to-rule-them-all

What it is

Task‑Observer (the one‑skill‑to‑rule‑them‑all) is a “skill” for Claude‑based AI agents. It runs alongside your normal Claude sessions, watches how you use the other skills you have installed, and records structured observations about:

• Corrections you make to an AI’s output
• Repetitive manual steps that could be turned into a new skill
• Gaps or blind‑spots in the observer itself These observations are turned into suggested updates for existing skills or proposals for brand‑new skills. The user reviews the suggestions and, once approved, the updated skill bundle is installed. The observer also logs its own improvements, making it a self‑refining meta‑skill.

Who might use it

• Knowledge‑workers who rely on Claude’s “skills” (custom extensions that add tools, prompts, or workflows) and want those extensions to evolve automatically.
• Builders of many Claude skills who need a systematic way to audit, improve, and generate new skills without manually editing each one.
• Anyone using Claude in a team or multi‑project setting where recurring friction should become reusable automation rather than ad‑hoc fixes.

How it works (high‑level)

1. Installation – Add the .skill bundle via the Vercel skills CLI (npx skills add …) or upload it in Claude’s UI.
2. Activation – Insert the activation snippet from references/environments.md into your CLAUDE.md (or equivalent) so the observer is invoked at session start.
3. Observation – During a session the skill watches tool calls, user corrections, and repeated patterns, writing a small JSON/YAML log to skill‑observations/observation‑log/ (or, on web‑only Claude, a hand‑off document).
4. Review – You periodically run a review session; the observer presents the logged items, you approve or reject them, and approved changes are staged in skill‑updates/.
5. Installation of updates – After review you install the updated skill bundles, completing the improvement cycle.

Key features

• Meta‑skill: monitors all other installed skills, including itself.
• Self‑improving: logs its own blind‑spots and can suggest changes to its own methodology.
• Cross‑cutting principles: extracts reusable patterns that apply to multiple skills.
• Human‑in‑the‑loop: no automatic overwrites; every suggestion must be approved.
• Platform‑specific handling: full filesystem logging on Claude Cowork/Code; hand‑off document mode on Claude web/mobile.

Installation & getting started

1. Grab the bundle – npx skills add rebelytics/one-skill-to-rule-them-all --skill task-observer (add -g for global install) or download the .skill zip from the latest release.
2. Activate – copy the snippet from references/environments.md into your CLAUDE.md (or the equivalent config file for your environment).
3. Verify – start a new Claude session and check that skill‑observations/observation‑log/ appears (or that a hand‑off doc is produced at session end).
4. Review cycle – schedule a recurring session (e.g., Mon/Wed/Fri) to approve logged observations and apply updates.

Ecosystem & compatibility

• Designed for Claude (Cowork, Dispatch, web, mobile, Claude Code). Works best where the skill can write to the filesystem.
• Community ports – a Codex version exists (AllstarGER) and users have reported integrations with Openclaw, Hermes, and other “skills‑compatible” platforms.
• License – CC‑BY‑4.0, allowing commercial use with attribution.

Where to learn more

• User guide: USER‑GUIDE.md in the repo.
• Methodology: the Augmented Expertise manifesto (https://www.rebelytics.com/augmented‑expertise/).
• Community listings – see the “Listed in” section for directories that index the skill.

---

All details above are taken directly from the repository’s README; no additional features have been inferred.