Zum Inhalt springen
TakoAPI
Zurück zu den Skills

agentmesh-governance

von imran-siddique

Coding Agents & IDEs
495Downloadzahlen laut clawskills.sh
0Aufrufe
0 LikesMelde dich an, um diesen Skill zu liken
—Noch nicht bewertet
Deine Bewertung

Kurzbeschreibung

AI agent governance, trust scoring, and policy enforcement powered by AgentMesh.

Beschreibung

Was dieser Skill macht

AgentMesh Governance adds a zero-trust layer to OpenClaw agents, enforcing token limits, tool allowlists/blocklists, and content policies before execution. Agents get Ed25519 cryptographic identities (DIDs) and trust scores across five dimensions. All actions are recorded in tamper-evident Merkle chain audit logs.

Unlike ad-hoc prompt-based restrictions, this enforces governance cryptographically and tracks trust scores over time, making it auditable and tamper-evident rather than easily bypassed.

Wann du ihn einsetzt

  • Blocking a research agent from running shell commands via policy file
  • Verifying another agent's DID before accepting its summarized output
  • Checking trust score before delegating a subtask to a writer agent
  • Proving execution integrity to a compliance team with Merkle-verified audit logs
  • Auto-blocking agents that repeatedly fail collaborative interactions

Beispiel-Workflow

EINGABE

User asks: enforce policy on a web_search action with 1500 tokens before executing

AGENT
  1. 1

    Run check-policy.sh with --action web_search --tokens 1500 --policy policy.yaml

  2. 2

    Governance engine evaluates action against max_tokens, allowed_tools, and blocked_patterns

  3. 3

    Returns JSON with allowed: true/false and any violations

  4. 4

    If allowed, proceed with tool execution

  5. 5

    Run record-interaction.sh after completion to update trust score

AUSGABE

JSON response: {"allowed": true, "violations": [], "recommendations": []}

Installation

clawhub install imran-siddique/agentmesh-governance

Führe den CLI-Befehl aus oder füge den Prompt in Claude Code, Cursor, Codex oder deinen Coding-Agenten ein.

Links