Skip to content

Task Observer - Setup Guide

Source: skills.sh (5.4K installs) GitHub: rebelytics/one-skill-to-rule-them-all (2,220⭐, CC BY 4.0) Category: Skill Lifecycle & Governance First Seen: Feb 18, 2026 (skills.sh) Quality Tier: 🟡 Trusted (Gen Agent Trust Hub Warn / Socket Pass / Snyk Pass)

Task Observer ("One Skill to Rule Them All") is a meta-skill by Eoghan Henn (rebelytics.com) that watches work sessions - autonomous or human - and converts friction into reusable skills. Instead of sitting down to "improve a skill," it formalizes the noticing: patterns, user corrections, workflow insights, and methodology get captured as one observation-log entry per insight, and proposed skill updates land in a review directory. The author's own log: 1,200+ observations across 70 skills in seven months, with most turned into skill improvements. The current version incorporates suggestions from 36 users across 60 issues and 12 pull requests. On the skills.sh hot board as of Sep 1, 2026, with steady 400-650 weekly installs.

Installation

npx skills add rebelytics/one-skill-to-rule-them-all --skill task-observer

Live-verified Sep 1, 2026: npx skills add rebelytics/one-skill-to-rule-them-all --list reports "Found 1 skill."

Hermes manual install (per the publisher's "other systems" guidance - keep the bundle intact wherever your platform expects skills):

git clone https://github.com/rebelytics/one-skill-to-rule-them-all.git /tmp/task-observer
mkdir -p ~/.hermes/profiles/corpusiq/skills/task-observer
cp -r /tmp/task-observer/SKILL.md /tmp/task-observer/references /tmp/task-observer/scripts \
  ~/.hermes/profiles/corpusiq/skills/task-observer/

The skill runs degraded without references/ and scripts/ - it will tell you which files are missing.

Prerequisites

Requirement Details
No API keys Fully local - observation logs live on your filesystem
Stable workspace path One absolute path that outlives sessions, pinned in the activation config (see Quick Start)
Session-start hook The SKILL.md states description matching alone is not enforceable - pair it with an AGENTS.md instruction or a harness session-start hook
Hermes Agent Uses the <available_skills> convention, which Hermes shares natively

What It Provides

Capability How It Works
Session observation Invoked before the first tool call of any multi-step session; watches for friction, corrections, and workflow patterns
Observation log One Markdown file with YAML frontmatter per insight at [workspace]/skill-observations/observation-log/; resolved entries archive under observation-log/archive/
Skill update proposals Proposed skill edits land in [workspace]/skill-updates/ for review
Skill authoring rules references/skill-authoring.md - includes a feedback pre-flight with duplicate checks across issues/PRs before submitting upstream
Signals & weekly review references/signals.md defines what counts as an observation; references/weekly-review.md drives the periodic consolidation pass
Bundle validation scripts/validate-skill-bundle.py checks the skill bundle's own integrity

Quick Start

  1. Install the skill (above)
  2. Pick a stable workspace path, e.g. ~/.hermes/profiles/corpusiq/vault/ - never a cwd inside an ephemeral checkout
  3. Add an activation instruction to the profile's AGENTS.md (or a session-start hook): invoke task-observer before the first tool call of any session
  4. Run real work. The observer writes observations as friction happens
  5. Review skill-updates/ periodically (or on the weekly-review cadence) and patch skills accordingly

CorpusIQ Use Cases

  • Quarterly skill audits - the observation log is the evidence layer: which skills hit friction in production ops, backed by dated entries instead of vibes
  • Skill sweep operations - sweeps like this one (skills.sh monitoring, docs maintenance) generate friction patterns the observer captures, feeding the improvement loop
  • Session post-mortems - corrections made during inbox, cron, and monitoring sessions become durable skill patches instead of being lost between sessions
  • Skill authoring feedback loop - the skill-authoring pre-flight (duplicate check across issues/PRs, upstream-HEAD verification) matches the patch-first discipline used for this catalog
  • Cross-agent consistency - the same observation discipline applies to dev@, support@, and growth agents, giving a unified improvement stream

Limitations / Verification

  • Gen Agent Trust Hub Warn caps the tier at 🟡 Trusted (Socket and Snyk both Pass) - the warning is named in this guide's Security section
  • Claude-architected content - the publisher's own words: the SKILL.md format is cross-platform but "the content assumes Claude's architecture" (Claude path conventions like ~/.claude/projects/, Cowork shared folders). Hermes shares the <available_skills> convention, but re-pin all workspace paths to a stable Hermes profile path; users have reported successful Hermes and OpenClaw integrations
  • Activation is the weak point - without the AGENTS.md/hook instruction, description matching alone will not reliably trigger the skill
  • CC BY 4.0 - share and adapt freely with attribution to Eoghan Henn / rebelytics.com
  • Verify install: npx skills list | grep task-observer (CLI) or ls ~/.hermes/profiles/corpusiq/skills/task-observer/ (manual)

Security

Gen Agent Trust Hub: Warn · Socket: Pass · Snyk: Pass

← Skills Catalog → Powered by CorpusIQ

Contact Terms Privacy Pricing