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

One‑Skill‑to‑Rule‑Them‑All (task‑observer)

What it is – A Claude skill (a small bundle of files) that acts as a “meta‑skill”. While you work with Claude‑based agents it watches every session, logs what it sees, and suggests new skills or improvements to existing ones – including improvements to itself. In practice it turns repetitive patterns, manual work, or corrections you make into concrete skill‑creation or update recommendations.

Why you might want it

• You already have a library of Claude skills and want them to keep evolving automatically.
• You build many skills and find manual auditing tedious.
• You prefer a systematic way to surface new skill ideas from everyday work.
• It works for non‑developers – the skill only produces suggestions; you approve the changes.

How it works

1. Observation – During a session the observer watches three signals:
   • Corrections you make to an AI’s output.
   • Repeated manual steps that could be automated.
   • Its own blind spots (e.g., missed observations).
2. Logging – It writes a structured observation log (one small file per observation) that records what was noticed, which skill(s) are affected, and concrete improvement ideas.
3. Review – You periodically review the logs, approve the suggested updates, and the corresponding skill files are edited (or new skills are created).
4. Self‑improvement – Because it also observes its own logs, the methodology it uses gets refined over time.

Key artefacts in the repo

• SKILL.md – the core definition of the task‑observer skill.
• references/ – supporting documentation (e.g., environment‑specific instructions).
• scripts/ – helper scripts used by the skill.
• USER‑GUIDE.md – step‑by‑step onboarding.
• CONTRIBUTING.md – how to report bugs, request features, or add compatibility reports.

Installation (Claude‑centric)

1. Download the .skill bundle from the latest release or clone the repo and keep SKILL.md, references/ and scripts/ together.
2. Claude web / desktop / mobile – upload the bundle via Settings → Customize or place the folder in ~/.claude/skills/task-observer/ (or the user‑level skills directory).
3. Add the activation instruction from references/environments.md to your CLAUDE.md (or equivalent) so the skill runs at session start.
4. Verify activation by checking that a folder skill‑observations/observation‑log/ appears after a few real work sessions.

Supported environments

• Claude Cowork (full filesystem experience)
• Claude Dispatch
• Claude.ai web interface (produces a hand‑off document instead of files)
• Claude mobile app (hand‑off document)
• Claude Code (desktop app)

The README notes community‑made ports for other platforms (e.g., a Codex version) and reports of successful use with Openclaw and Hermes, though those require custom adaptation.

Who should try it

• Knowledge workers who already use Claude skills and want them to evolve automatically.
• Builders of many skills who need a systematic audit/review loop.
• Anyone new to skills who wants the system to suggest useful skill ideas based on their workflow.

Contribution & community

• Open‑source under CC‑BY‑4.0 – you can remix, commercialise, or redistribute with attribution.
• Contributions are welcomed via issues or pull requests; the project already incorporates feedback from 50 users, 103 issues, and 26 PRs.
• The maintainer encourages reports of compatibility with other AI platforms and creative use‑case stories.

Where to learn more

• User guide: USER-GUIDE.md (linked in the README)
• Methodology: the Augmented Expertise manifesto on rebelytics.com
• Security audit badge from Oathe Security (linked in the README).

---

Created by Eoghan Henn (rebelytics.com). All information above is taken directly from the repository’s README.