One Skill to Rule Them All (task-observer): a meta-skill that watches your work and improves your other skills
The meta-skill that builds and improves all your skills, including itself. Watches your work sessions (autonomous or human-led), captures patterns, corrections and judgement calls, and turns them into skill improvements and new skill candidates for your review. Practical application of the Augmented Expertise methodology. Open source: CC BY 4.0.
At a glance
- What is it?
- task-observer by Rebelytics is a CC BY 4.0 meta-skill for Claude and other agents that runs alongside your sessions, captures corrections, patterns and gaps, and turns them into review-ready suggestions for improving existing skills and creating new ones. It observes itself too, and never edits your skills without your approval.
- Who is it for?
- Adopt task-observer if you maintain several Claude skills and want them to improve from real use instead of staying frozen. Grab the bundle, keep SKILL.md, references/ and scripts/ together, and, crucially, add the activation instruction to your CLAUDE.md and verify in a new session that observation logs appear.
- Can I use it commercially?
- Yes, with credit. CC-BY-4.0 allows commercial use as long as you credit the authors and indicate what you changed. It is written for creative content, so check how it applies to any code.
- Is it still maintained?
- Yes. The repository last received commits 4 days ago.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What task-observer is
task-observer, styled One Skill to Rule Them All, is a meta-skill: a skill whose job is to build and improve all your other skills, including itself. It runs alongside your work, watches what you do, and does two things. It identifies new skills by spotting repeating patterns and flagging them as candidates, then helps you build them; and it improves existing skills by noticing your corrections, stated preferences and gaps, then suggesting specific updates. You work normally, it watches, and your skill library grows and improves over time.
The README grounds this in the author's own use: in seven months it logged almost 1400 observations across 78 skills, most of which became improvements, and the majority of those 78 skills were themselves created from its observations. The current version also folds in contributions from 50 users across 103 issues and 26 pull requests. It is a practical application of Rebelytics' Augmented Expertise methodology, but users report using it with Hermes and OpenClaw too, so it is not Claude-only.
The three things it watches for
During each session, task-observer looks for three signals. First, corrections and adjustments: when you steer or edit the AI's output, that is evidence a skill could be clearer or more complete. Second, gaps no skill covers yet: when you do something manually or repeatedly that could be systematised, it flags a candidate for a new skill. Third, its own blind spots: because it observes all active skills including itself, it captures improvements to its own methodology, notices when its observation format is unclear, and notices when it triggers in the wrong contexts.
Each session produces a structured observation log: what it noticed, which skills are affected, and specific suggested improvements. Some observations are not specific to one skill; these become cross-cutting principles in a separate log, and new skills are automatically checked against them when created or updated, which the README argues raises the quality floor across the whole library over time.
You stay in control
A design choice worth emphasising is that the observer does not modify your skills directly. It produces recommendations that you review, approve or reject, so you decide what changes and when. That matters for trust: a tool that rewrites your skills automatically would be risky, whereas one that proposes changes for review is a safe addition to an existing setup.
The self-improving loop is the part the README is most enthusiastic about, and it is genuinely the interesting idea: the skill that improves all your skills also improves itself, because it runs during every session and observes itself alongside everything else. Over time that means both your skills and the observer's own methodology get sharper, without you auditing each skill by hand.
Installing the bundle
The skill is a small bundle: SKILL.md, on-demand files in references/ (kept separate so the always-loaded part stays lean), and two helper scripts in scripts/. The README is explicit that installing only SKILL.md runs in a degraded mode and is not recommended, so keep the three parts together.
To get the files, download the .skill bundle from the latest release, or download the repo as a ZIP or clone it, keeping SKILL.md, references/ and scripts/ together. For the Claude web, desktop, mobile or Cowork apps, upload the .skill bundle via Settings then Customize, or zip a folder containing the three parts. For Claude Code, place the folder at .claude/skills/task-observer/ (project level) or in your user-level skills directory, preserving the subfolders. For other systems, keep the folder structure intact wherever your platform expects skills and let your AI guide the setup.
The activation step most people miss
The README's most important practical warning is that installing the files is not the same as activating the skill. Description matching alone under-triggers, so you must add the activation instruction from references/environments.md to your CLAUDE.md (or your platform's equivalent), or install the session-start hook.
Then verify it in a new session, because the session you install in cannot prove activation. The external tell is concrete: if skill-observations/observation-log/ does not exist after a few sessions of real work, activation never happened. This is the single step that determines whether the tool does anything at all, and it is easy to skip, so treat the verification, seeing an observation log appear in a fresh session before the first tool call, as part of installation rather than an optional check. In Claude Cowork or Claude Code in the desktop app, the observer writes logs to your filesystem, so improvements persist between sessions.
Who it suits, licence and an honest boundary
task-observer is aimed at anyone who uses skills and wants them to improve rather than stay frozen; you do not need to be a developer. It is particularly valuable if you have built multiple skills and want a systematic way to maintain them without auditing each one, and it can also help people with no skills yet by identifying candidates to build.
The README states an honest boundary, which is a good sign of a project that understands its own value. The formal observation log and review cycle pay off most as your library and usage grow, many skills, parallel sessions, scheduled reviews. For a small setup with a handful of skills, your AI's built-in memory may cover much of the same ground with less overhead, and editing a skill directly is quick. The value compounds with scale, so adopt it early if you expect growth, or return to it when direct editing stops feeling manageable. The project is open source under CC BY 4.0, with recent releases v3.2.0, v3.1.0 and v3.0.0 through late August and September and the last push on 2026-09-11.
Editorial conclusion
Adopt task-observer if you maintain several Claude skills and want them to improve from real use instead of staying frozen. Grab the bundle, keep SKILL.md, references/ and scripts/ together, and, crucially, add the activation instruction to your CLAUDE.md and verify in a new session that observation logs appear. Its value compounds with scale, so it pays off most once your skill library is large.
Frequently asked questions
What is task-observer (One Skill to Rule Them All)?
It is a meta-skill that runs alongside your Claude sessions, watches your work, and turns corrections, repeated patterns and gaps into review-ready suggestions to improve your existing skills and create new ones. It observes itself too and never edits skills without your approval.
How do I make sure it is actually running?
Installing the files is not enough. Add the activation instruction from references/environments.md to your CLAUDE.md or install the session-start hook, then verify in a new session. If skill-observations/observation-log/ does not appear after a few real sessions, activation did not happen.
Is it worth it if I only have a few skills?
The README is honest that its value compounds with scale. For a small setup, your AI's built-in memory may cover similar ground with less overhead and direct editing is quick. Adopt it early if you expect your skill library to grow, or return to it when manual editing stops being manageable.
Community notes