Task Observer — One Skill to Rule Them All
A meta-skill that watches your work sessions and logs friction, corrections and reusable methodology as durable observations for continuous skill improvement.
AutomationAdvanced★ 3,130⑂ 302AI score 9/10Last updated: Oct 2, 2026
What it does
- Runs as a background observer across any multi-step task or agentic workflow, capturing user corrections, recurring task types, and "I always do it this way" methodology that no existing skill encodes.
- Stores each finding as one Markdown file with YAML frontmatter under
skill-observations/observation-log/, anchored on a single stable absolute workspace path that outlives individual sessions (with anarchive/for resolved entries). - Ships a strict Session Start Protocol: probe the storage mount inside the first batched tool call, scan frontmatter only (cheap at hundreds of entries), decide whether a weekly review is due, verify the activation config, and reconcile staged skill updates.
- Enforcement is the point: mandatory disk writes after every 3rd completed todo, a flush at every deliverable hand-off, an immediate flush when an external write is denied, plus retry-once-and-try-another-tool rules so observations are never silently lost.
- Includes de-duplication and escalation rules — extend an existing open entry instead of restating it, and when a rule is violated a second time, propose a structural barrier (hook, lint rule, unavailable default) rather than clearer wording.
Who it's for
- Builders who maintain their own set of Claude Code skills and want them to get measurably better over time.
- Anyone running long agent sessions who hates losing hard-won lessons at session end.
- Tech leads documenting and standardising a team's working methodology.
- Probably overkill if you just want a small, self-contained skill: the core file is large and demands careful path setup.
Examples
- You correct the agent three times to "run the tests first" — an observation is logged that the target skill lacks that rule, and it surfaces in the next review.
- The agent breaks the same documented rule twice — the proposal must now be a hook that refuses the call, not a bolder warning.
- First run with an empty log but a repo full of handover docs and commit history — the skill offers a one-off backfill pass to seed initial observations.
· · · Install guide · · ·
Try it now, no install
Paste this into Claude to use the skill without installing anything.
Read the instructions in this file and follow them to help me: https://raw.githubusercontent.com/rebelytics/one-skill-to-rule-them-all/HEAD/SKILL.md What I want: (describe your task here)
If Claude can't open the link, open it yourself and paste the contents instead.
↓ If it works for you, download the ZIP below and install it. Then it runs on its own — no pasting each time.
Install in the Claude app (no terminal)
- Download the ZIP with the button below.
- In Claude, open Settings → Capabilities and turn on 'Code execution and file creation'. (one time)
- Go to Customize → Skills → + → 'Upload a skill' and upload the ZIP.
Install in Claude Code
Let Claude do it — paste this into Claude Code
Install the skill I found on Claude Skill Mart. Copy the SKILL.m folder from the GitHub repo rebelytics/one-skill-to-rule-them-all into my ~/.claude/skills/task-observer/. When it's done, tell me in one line what this skill can do.
Install with a command instead
git clone https://github.com/rebelytics/one-skill-to-rule-them-all.git ~/.claude/skills/task-observer⚠ This is a third-party skill. Check the source repository before installing.
- Open a terminal.
- Make sure the skills folder exists:
mkdir -p ~/.claude/skills - Clone the repository:
git clone https://github.com/rebelytics/one-skill-to-rule-them-all.git ~/.claude/skills/task-observer - Confirm you got
SKILL.mdand the fullreferences/bundle (weekly-review, skill-authoring, observation-log, signals, environments, migration, starter-principles). A missing reference file means that episode will not run. - Pick ONE stable absolute path for the observation workspace, e.g.
~/.claude/skill-observations. Prefer a single user-scope path so the log is not sharded per project, and never put it inside an ephemeral git worktree or throwaway clone. - Substitute every
[ABSOLUTE PATH]in the snippets with that path. Keep it double-quoted (it may contain a space) and run the snippets withbash, notsh. - Add an activation instruction to your
CLAUDE.mdor a session-start hook: invoke task-observer before the first tool call of a session. Description matching alone is not enforceable. - Restart Claude Code and say "One Skill to Rule Them All" or "skill observations" to verify it activates and creates the workspace artefacts.
View source on GitHub ↗License: CC-BY-4.0