gatewaystack-governance

Coding Agents & IDEs
v0.2.0
Benign

Deny-by-default governance for every tool call — identity, scope, rate limiting, injection detection, audit.

2498 downloads498 installsby @davidcrowe

Setup & Installation

Install command

clawhub install davidcrowe/gatewaystack-governance

If the CLI is not installed:

Install command

npx clawhub@latest install davidcrowe/gatewaystack-governance

Or install with OpenClaw CLI:

Install command

openclaw skills install davidcrowe/gatewaystack-governance

or paste the repo link into your assistant's chat

Install command

https://github.com/openclaw/skills/tree/main/skills/davidcrowe/gatewaystack-governance

What This Skill Does

Deny-by-default governance layer for OpenClaw that intercepts every tool call before execution. Runs five mandatory checks: identity verification, scope enforcement, rate limiting, prompt injection detection, and audit logging. Three optional features add output DLP, human escalation, and behavioral monitoring.

Because it hooks at the process level, the agent cannot route around it the way it could with in-prompt or middleware-based guardrails.

When to Use It

  • Blocking an agent from calling tools outside its approved scope
  • Detecting prompt injection attempts before a tool executes
  • Reviewing audit logs after an unexpected agent action
  • Redacting PII from tool output before it reaches the agent
  • Triggering human review when an agent tries a tool for the first time
View original SKILL.md file
# GatewayStack Governance

Deny-by-default governance for every tool call in OpenClaw.

Five core checks run automatically on every invocation:

1. **Identity** — maps the agent to a policy role. Unknown agents are denied.
2. **Scope** — deny-by-default tool allowlist. Unlisted tools are blocked.
3. **Rate limiting** — per-user and per-session sliding window limits.
4. **Injection detection** — 40+ patterns from Cisco, Snyk, and Kaspersky research.
5. **Audit logging** — every decision recorded to append-only JSONL.

Three opt-in features extend governance further:

6. **Output DLP** — scans tool output for PII using `@gatewaystack/transformabl-core`. Log or redact.
7. **Escalation** — human-in-the-loop review for medium-severity detections and first-time tool use.
8. **Behavioral monitoring** — detects anomalous tool usage patterns using `@gatewaystack/limitabl-core`.

## Install

```bash
openclaw plugins install @gatewaystack/gatewaystack-governance
```

One command. Zero config. The core 5 checks are active on every tool call immediately.

The plugin hooks into `before_tool_call` at the process level — the agent can't bypass it, skip it, or talk its way around it.

## Customize

To override the defaults, create a policy file:

```bash
cp ~/.openclaw/plugins/gatewaystack-governance/policy.example.json \
   ~/.openclaw/plugins/gatewaystack-governance/policy.json
```

Configure which tools are allowed, who can use them, rate limits, injection detection sensitivity, and the three optional features (DLP, escalation, behavioral monitoring — all disabled by default).

## Optional GatewayStack packages

The opt-in features use GatewayStack packages via lazy import. Install only what you need:

```bash
npm install @gatewaystack/transformabl-core   # for output DLP
npm install @gatewaystack/limitabl-core       # for behavioral monitoring
```

The core 5 checks have zero external dependencies and work without these packages.

## Links

- [GitHub](https://github.com/davidcrowe/openclaw-gatewaystack-governance) — source, docs, getting started guide
- [npm](https://www.npmjs.com/package/@gatewaystack/gatewaystack-governance) — package registry
- MIT licensed

Example Workflow

Here's how your AI assistant might use this skill in practice.

INPUT

User asks: Blocking an agent from calling tools outside its approved scope

AGENT
  1. 1Blocking an agent from calling tools outside its approved scope
  2. 2Detecting prompt injection attempts before a tool executes
  3. 3Reviewing audit logs after an unexpected agent action
  4. 4Redacting PII from tool output before it reaches the agent
  5. 5Triggering human review when an agent tries a tool for the first time
OUTPUT
Deny-by-default governance for every tool call — identity, scope, rate limiting, injection detection, audit.

Share this skill

Security Audits

VirusTotalBenign
OpenClawBenign
View full report

These signals reflect official OpenClaw status values. A Suspicious status means the skill should be used with extra caution.

Details

LanguageMarkdown
Last updatedFeb 25, 2026