Skip to content

trendinghot/clawshield

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 

Repository files navigation

# ClawShield

**Governance sidecar for OpenExec**

ClawShield exists to make autonomous execution **legible**.

It does not run tasks.

It does not modify behavior.

It does not fix mistakes.

ClawShield defines **policy boundaries**, evaluates actions **before execution**, and produces **verifiable receipts** of what was allowed or denied.

When systems act, ClawShield makes those actions **explainable**.

This matters when actions must be reviewed, shared, or defended outside the system that performed them.


System architecture

This project is part of a three-layer governed execution system:

  • OpenExec executes actions deterministically.
  • ClawShield governs whether actions are permitted before execution.
  • ClawLedger records immutable receipts of what was decided and why.

Execution, governance, and evidence are intentionally separated. No layer replaces another. Together, they make autonomous systems legible, reviewable, and defensible.

## What you are looking at

If you continue, you will enter a **governance workspace** containing:

- **Policies** that define what an agent *may* attempt

- **Preflight checks** that evaluate actions before execution

- **Receipts** that record governance decisions immutably

Nothing runs automatically.

No background enforcement is active by default.

ClawShield governs decisions.

Other systems execute them.


## What ClawShield is not

ClawShield is **not**:

- An execution engine

- An automation framework

- An AI agent

- A monitoring or surveillance system

It does not schedule work.

It does not retry failures.

It does not intervene after execution begins.

ClawShield draws boundaries.

It does not cross them.


## Governance model

ClawShield operates in **default-deny** mode.

Every action is evaluated explicitly against policy.

If a policy does not allow it, the action is denied.

A denial is not an error.

It is a boundary being enforced.

Each decision produces a **sealed receipt** containing:

- The evaluated action

- The target

- The policy hash

- The decision (allowed or denied)

- A timestamp

- A sealed marker indicating immutability

Receipts are evidence.

They do not expire.


## Relationship to OpenExec

OpenExec executes tasks.

ClawShield governs whether those tasks *should be executed*.

ClawShield does not fork OpenExec.

It does not wrap execution.

It does not intercept runtime behavior.

It evaluates intent **before execution** and records the decision.


## Modes

### Free — Learning Governance

- Manual preflight checks

- View receipts

- Learn policy boundaries

- Up to 5 active policies

### Governed — Operating with Responsibility

- Export evidence

- Continuous enforcement

- Programmatic access

- Unlimited policies

- Policy templates

Governed mode does not make decisions safer.

It makes decisions **survivable**.


## Design principles

- Governance must be **explicit**

- Boundaries must be **visible**

- Decisions must be **provable**

- Evidence must be **portable**

- Silence is preferable to hype

If ClawShield blocks something you did not expect, that is a signal — not a failure.


## Status

This repository is the **canonical reference** for ClawShield’s governance model.

The running application and UI are implemented elsewhere.

This repository defines intent, scope, and guarantees.


## License

To be defined.


Related repositories

About

Governance sidecar for OpenClaw

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published