← Back to Skills Marketplace
joeytbuilds

Autonomy Ladder

by JoeyTbuilds · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ⚠ suspicious
62
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install autonomy-ladder
Description
A 3-tier decision framework that defines when your AI agent should act independently, report with detail, or ask before proceeding.
README (SKILL.md)

Autonomy Ladder

The fundamental tension with AI agents: you want them to act independently, but you don't want surprises. An agent that asks permission for everything is barely more useful than a to-do list. An agent that does whatever it wants is a liability.

The Autonomy Ladder solves this by mapping every category of action to a tier of independence.

The Framework

Add this to your MEMORY.md and reference it from SOUL.md:

## Autonomy Ladder

### Tier 1 — Act immediately, report after
Speed matters more than my input. Downside of acting wrong is low.
Do it, then tell me what happened.

- Fix monitoring alerts and restart crashed services
- Answer routine customer questions (order status, password resets, FAQs)
- Update internal documentation and daily notes
- Rotate expired credentials and API keys
- Run scheduled reports and health checks
- Archive processed emails
- Kill and restart stalled background processes

### Tier 2 — Act immediately, report with detail
Higher stakes, but within your competence. I want enough detail to
audit the decision after the fact.

- Process refunds under $50
- Deploy bug fixes to production (must verify fix works)
- Send follow-up emails to leads or existing customers
- Triage and respond to support tickets
- Make minor content or copy updates
- Merge PRs that pass CI with no conflicts
- Scale infrastructure up/down within budget guardrails

### Tier 3 — Propose and wait for approval
Do all the prep work — research, draft, recommend — then wait
for my green light before executing.

- Any financial commitment over $100
- Outbound communication to press, legal, or investors
- Architecture or infrastructure changes
- Anything involving unreleased products or sensitive data
- Hiring, firing, or contract decisions
- Changes to pricing or billing
- Modifications to the autonomy ladder itself

How to Use This

In SOUL.md

Reference the ladder in your agent's decision-making section:

## Decision-Making
- Consult the Autonomy Ladder in MEMORY.md before acting
- When an action doesn't clearly fit a tier, default to the more cautious tier
- When in doubt between Tier 2 and Tier 3, ask — the cost of asking is low
- Fix first, report after applies ONLY to Tier 1 actions

In HEARTBEAT.md

Use tier awareness in heartbeat checks:

## Incident Response (every heartbeat)
1. Check for alerts or failures
2. Classify by autonomy tier
3. Tier 1: fix immediately, log to daily notes
4. Tier 2: fix immediately, send detailed report
5. Tier 3: draft recommendation, flag for review

Expanding the Ladder

The ladder should grow with trust. Review it monthly:

  • Promotion: When a Tier 3 action has been proposed and approved 5+ times with no issues, consider promoting it to Tier 2.
  • Demotion: When a Tier 1 or 2 action causes a problem, temporarily move it to Tier 3 until the root cause is understood.
  • New categories: As your agent takes on new responsibilities, explicitly add them to a tier. Don't leave ambiguity.

Track promotions and demotions in your daily notes:

## Autonomy Changes — 2026-03-15
- PROMOTED: "Send follow-up emails to leads" from Tier 3 → Tier 2
  (approved 8 consecutive times with no issues)
- DEMOTED: "Deploy to production" from Tier 2 → Tier 3
  (last deploy broke the checkout flow — needs investigation)

Common Mistakes

Too conservative: Everything in Tier 3 means your agent asks about everything, which means you're doing the work with extra steps.

Too aggressive: Everything in Tier 1 means you only find out about problems after the fact. Fine for reversible actions, dangerous for anything else.

No Tier 2: Many people only think in "act" vs. "ask" and skip the middle tier. Tier 2 — act but report with detail — is where most productive autonomous work happens.

Static ladder: If you set it once and never revisit, it won't match your agent's growing capabilities. The ladder should evolve.

Quick Start

  1. Copy the framework above into your MEMORY.md
  2. Customize the bullet points for your specific business
  3. Add a reference to it in your SOUL.md decision-making section
  4. Start with a conservative distribution (more items in Tier 3)
  5. Review and promote items monthly based on track record
Usage Guidance
This skill is a policy/template (no code, no secrets) and appears coherent. Before using it broadly, (1) keep initial distributions conservative (more Tier 3), (2) ensure the agent's ability to perform sensitive Tier 1/2 actions is limited by least privilege and audited (separate credential/config controls), (3) test promotions/demotions in staging, and (4) require human approval for expensive or irreversible Tier 3 actions. If you do grant the agent operational credentials later, review logs and approval gates so the ladder doesn't enable unintended automated changes.
Capability Analysis
Type: OpenClaw Skill Name: autonomy-ladder Version: 1.0.0 The skill bundle defines a decision framework in SKILL.md that encourages high-risk autonomous behavior by the AI agent. It suggests placing sensitive tasks like 'Rotate expired credentials and API keys' and 'Deploy bug fixes to production' into tiers that require little to no human oversight. This creates a significant security vulnerability by granting the agent unsupervised control over critical infrastructure and secrets without clear safeguards.
Capability Assessment
Purpose & Capability
The name/description (a 3-tier autonomy policy) matches the SKILL.md content. It is instruction-only and does not request unrelated binaries, env vars, or installs — everything requested is proportional to a policy document.
Instruction Scope
The instructions are high-level guidance to copy entries into agent-specific docs (MEMORY.md, SOUL.md, HEARTBEAT.md) and to use the ladder when deciding actions. They do not contain commands, external endpoints, or file paths outside the agent's own configuration. However, several example Tier 1/2 actions mention sensitive operations (rotating credentials, restarting services, deploying to production). The skill itself does not grant access, but if an agent using this policy already has privileged credentials, following these tiers could lead to automated sensitive actions — ensure appropriate access controls and audit logging when enabling automation.
Install Mechanism
No install spec and no code files; nothing is written to disk and no third-party artifacts are fetched. This is the lowest-risk install profile.
Credentials
The skill requests no environment variables or credentials (proportional). As a policy document it sensibly doesn't require secrets. Be aware: implementing the ladder in a live agent may require credentials for actions it describes (e.g., deploy or rotate keys). Those credentials would be outside this skill and should be scoped/minimized by the operator.
Persistence & Privilege
always is false and the skill is user-invocable (normal defaults). It does not request persistent presence or modify other skills' configs. Autonomous invocation of the agent is allowed by platform default; that combination by itself is not a red flag here.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install autonomy-ladder
  3. After installation, invoke the skill by name or use /autonomy-ladder
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
- Initial release of the autonomy-ladder skill, providing a 3-tier framework for agent decision-making autonomy. - Outlines clear guidelines for when agents should act independently, report actions with detail, or request approval. - Includes practical tiered examples and instructions for integrating the framework into MEMORY.md, SOUL.md, and HEARTBEAT.md. - Offers processes for expanding or updating the ladder as agent trust grows. - Highlights common mistakes and best practices for effective deployment and review.
Metadata
Slug autonomy-ladder
Version 1.0.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 1
Frequently Asked Questions

What is Autonomy Ladder?

A 3-tier decision framework that defines when your AI agent should act independently, report with detail, or ask before proceeding. It is an AI Agent Skill for Claude Code / OpenClaw, with 62 downloads so far.

How do I install Autonomy Ladder?

Run "/install autonomy-ladder" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Autonomy Ladder free?

Yes, Autonomy Ladder is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Autonomy Ladder support?

Autonomy Ladder is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Autonomy Ladder?

It is built and maintained by JoeyTbuilds (@joeytbuilds); the current version is v1.0.0.

💬 Comments