Skip to main content
Version: v1.1.0
This page describes an integration pattern. It does not change TealTiger v1.1.0 contracts.

​
TealTiger + OpenClaw

​
Why this integration exists

OpenClaw is a local-first autonomous agent that can interact with real systems (messages, email, files, tools). This creates powerful automation — and a large blast radius if actions are misconfigured or manipulated. TealTiger adds runtime governance semantics to OpenClaw actions:
  • Deterministic Decisions (what to do)
  • Reason Codes (why)
  • Risk Scores (how risky)
  • Audit Evidence (what happened)
  • Enforcement Modes (report/monitor/enforce)

​
Important note: hardening vs governance

Some OpenClaw tools focus on security hardening (auditing configs, tightening gateway exposure, credential hygiene, supply-chain checks). These reduce the chance of compromise. TealTiger focuses on runtime governance: whether a specific action should be allowed right now, with deterministic decisions and audit evidence.
Use hardening tools to reduce attack surface, and use TealTiger to govern high-impact actions with policy and evidence.
See also: /concepts/security-vs-governance

​
Integration goal (v1.1.0)

Introduce a deterministic policy check before and/or after OpenClaw performs high-impact actions such as:
  • tool/skill invocation
  • filesystem access
  • shell/command execution
  • network calls
  • sending messages/emails

​
Two supported integration models

​
Model A — OpenClaw Plugin Hook (in-process)

A small plugin calls TealTiger before executing an action and returns a Decision. Best for:
  • lowest latency
  • simplest deployment
Tradeoffs:
  • plugin boundary is privileged; treat plugin code as trusted.

​
Model B — Local Sidecar (out-of-process)

OpenClaw (or a plugin) calls a local TealTiger evaluator over localhost. Best for:
  • stronger isolation
  • easier enterprise review
Tradeoffs:
  • slightly more plumbing

​
What TealTiger evaluates (RequestContext for OpenClaw)

At minimum, the integration should supply:
  • identity (agent/user/channel)
  • environment (prod/dev)
  • action intent (read/write/exec/message/network)
  • tool metadata (name + normalized args)
  • correlation (trace_id/span_id/request_id)

​
Decision mapping to OpenClaw behavior

  • ALLOW → execute
  • DENY → block
  • REDACT → redact then proceed
  • TRANSFORM → deterministic modifications
  • DEGRADE → reduced capability
  • REQUIRE_APPROVAL → defer

​
Enforcement Modes

  • REPORT_ONLY: allow, record evidence
  • MONITOR: allow, record violations
  • ENFORCE: block or defer

​
Audit Evidence

TealTiger emits structured audit events capturing decision outcome, reason codes, optional risk scores, redaction status, and correlation identifiers.

​
Next steps

  • Start with REPORT_ONLY.
  • Add minimal policies (deny obviously dangerous actions first).
  • Export audit events to your logging/SIEM pipeline.
  • Move to MONITOR then ENFORCE.
  • /concepts/decision-model
  • /policy/reason-codes
  • /policy/risk-scores
  • /audit/audit-event-schema
  • /architecture/enforcement-flow
  • /concepts/security-vs-governance