DestructGuard Command Score

$0.10/call

Catch destructive agent commands before they run.

POST /api/agent-services/destructguard/score

agent-safetydestructive-actionsx402destructguard

What it does

Deterministic risk scorer for autonomous-agent commands. Scores a shell, SQL, or git action for destructive, deployment, and data-mutation risk before execution and returns a risk level, the matched policy rules, and an approve-or-block recommendation. Pure rules, evidence per match, no fabrication.

Scores shell / SQL / git actions for risk
Flags force-push, hard-reset, drops, teardown
Returns an approve-or-block recommendation

Example request

POST /api/agent-services/destructguard/score

{
  "command": "rm -rf /",
  "repo": "github.com/acme/app",
  "cwd": "/workspace/app",
  "actor": "codex",
  "intent": "cleanup generated files",
  "diffStat": "3 files changed, 18 insertions(+), 44 deletions(-)"
}

Example response (HTTP 200)

Deterministic — the same inputs always return the same audited output.

{
  "status": "scored",
  "service": "destructguard-command-score",
  "requestId": "dg_<uuid>",
  "risk": "critical",
  "requiresApproval": true,
  "recommendedAction": "block_until_explicit_human_or_policy_approval"
}

Input schema

Top-level request fields. Nested shapes are shown in the example above and the OpenAPI spec.

Field	Type	Required	Description
command	string	Yes	Shell command or agent action to score
repo	string	—
cwd	string	—
actor	string	—
intent	string	—
diffStat	string	—

How to call it over x402

1. Send the request. The first unpaid call returns HTTP 402 with an x402 payment challenge — $0.10, USDC on Base, and the recipient.
2. Pay per call. Your x402 client signs the USDC payment and retries automatically — no API key, no account, no subscription. New to x402?
3. Read the result. HTTP 200 returns the computed values plus evidence-backed findings.

With the x402 fetch client (Node / TypeScript)

import { wrapFetchWithPayment } from "@x402/fetch";
import { privateKeyToAccount } from "viem/accounts";

const account = privateKeyToAccount(process.env.AGENT_WALLET_KEY);
const pay = wrapFetchWithPayment(fetch, account); // USDC on Base

const res = await pay("https://hermesplant.com/api/agent-services/destructguard/score", {
  method: "POST",
  headers: { "content-type": "application/json" },
  body: JSON.stringify({
  "command": "rm -rf /",
  "repo": "github.com/acme/app",
  "cwd": "/workspace/app",
  "actor": "codex",
  "intent": "cleanup generated files",
  "diffStat": "3 files changed, 18 insertions(+), 44 deletions(-)"
}),
});

const result = await res.json();

Inspect the 402 with curl

curl -i -X POST https://hermesplant.com/api/agent-services/destructguard/score \
  -H "content-type: application/json" \
  -d '{"command":"rm -rf /","repo":"github.com/acme/app","cwd":"/workspace/app","actor":"codex","intent":"cleanup generated files","diffStat":"3 files changed, 18 insertions(+), 44 deletions(-)"}'
# → HTTP/1.1 402 Payment Required  (x402 challenge: price, USDC asset, Base network, recipient)
# → sign the USDC-on-Base payment and retry to receive HTTP 200

Prefer zero code? This endpoint is also exposed as a tool on the Hermes Plant MCP server, so an MCP-capable agent can call it with its own x402 wallet.