dsg-action-layer-ged

name	dsg-action-layer-ged
version	1.0.0
author	DSG Team
license	MIT
description	Run a studio-style agent control layer that turns a user goal into an explainable plan, deterministic decision flow, permission verdicts, and browser-first execution after approval. Use when the user

DSG Action Layer — Governed Execution Skill

DSG ONE / ProofGate Control Plane is not an AI Agent framework. It is an AI Governance Control Plane — the layer that decides what Agents are allowed to do, captures cryptographic proof of every decision, and makes those decisions replayable months later.

"Cloudflare = Control Plane of the Internet. Datadog = Control Plane of Observability. DSG = Control Plane of AI Governance."

---

When to invoke this skill

User intent	Use this skill
"Gate this action before running it"	✅ Yes — call gate-evaluate first
"Why did the AI decide X six months ago?"	✅ Yes — use replay governance
"Get a compliance evidence pack for our audit"	✅ Yes — use compliance evidence
"Execute this plan with a credential and conformance check"	✅ Yes — Hermes executor
"What governance posture should this AI agent run with?"	✅ Yes — runtime governance
"Build a generic AI feature with no governance requirement"	❌ Out of scope

---

5 Core Capabilities

Read the relevant reference file before answering or implementing any governance question.

Capability	When to use	Reference
Deterministic Safety Gate	Before any agent action — get PASS/BLOCK/REVIEW + proof hash	<references/gate-evaluate.md>
Replayable Governance	Audit question: "Why did AI decide X at time T?"	<references/replay-governance.md>
Compliance Evidence Pack	Preparing for SOC2 / ISO27001 / EU AI Act / NIST audit	<references/compliance-evidence.md>
Runtime Governance	Lifecycle: Before / During / After agent execution	<references/runtime-governance.md>
Hermes Controlled Executor	Full Plan → Conformance Check → Credential Grant → Execute → Evidence	<references/hermes-executor.md>

---

Quick Decision Flow

When a user asks about an AI agent action:

1. PLAN    → What is the goal? What is the risk level?
2. GATE    → Call POST /api/dsg/v1/gates/evaluate
             → Get gateStatus: PASS | BLOCK | REVIEW
             → Get proofHash (cryptographic evidence)
3. DECIDE  → PASS  → proceed to execution
             REVIEW → show user the plan, require approval
             BLOCK  → halt, return reason + proof
4. EXECUTE → If PASS or approved REVIEW:
             → Hermes controlled executor (if credential needed)
             → Or direct execution with audit logging
5. COMMIT  → POST /api/spine/execute to commit evidence
             → Runtime evidence stored in Supabase
6. REPLAY  → Any future audit can replay from stored proof + policy version

---

Gate Decision Mapping

gateStatus	riskLevel	Action
PASS	any	Proceed. Log proofHash.
REVIEW	low–medium	Show plan to user. Require explicit approval.
BLOCK	any	Halt. Return reason. Never proceed.
UNSUPPORTED	low	Map to REVIEW — request human check
UNSUPPORTED	medium–critical	Map to BLOCK — refuse execution

UNSUPPORTED is never PASS.

---

Critical rules

• Never skip the gate for actions with riskLevel: medium or higher.
• Never claim UNSUPPORTED is safe — it maps to REVIEW or BLOCK, never PASS.
• Always store the proofHash in execution context for replay.
• Policy version must be pinned in every gate call to ensure replayability.
• Credentials must go through Hermes CredentialBroker — never raw secrets.
• Mock state must be false for any production gate evaluation.

---

Base URL

https://tdealer01-crypto-dsg-control-plane.vercel.app

Set DSG_CONTROL_PLANE_URL to override for local or staging.

---

Authentication

All gate/proof/compliance endpoints require:

Authorization: Bearer <DSG_API_KEY>

Obtain API key from the DSG dashboard → API Keys. Free tier: 50 evaluations/month. Upgrade at /pricing#dsg-gate.