ScopeBlind Passport — Signed Access Control for Your Agent

What This Skill Does

This skill wraps your OpenClaw agent's MCP tool calls through protect-mcp, adding:

• - Shadow mode — logs every tool call with a signed receipt (blocks nothing by default)
• Per-tool policies — block, rate-limit, or require approval for specific tools
• Signed receipts — Ed25519-signed, JCS-canonicalized proof of every decision
• Trust tiers — available for advanced configurations to gate tool access by verified track record
• Local daily digest — human-readable summary of what your agent did

Any platform can log what its agents do. Very few will let you verify those logs without trusting them.

When to Use This Skill

Use this skill when:

• - You want to know what your agent did while you weren't watching
• You need signed, tamper-proof proof of agent actions for compliance or auditing
• You want to block dangerous tools (deletedatabase, sendemailasceo, rm_rf)
• You want rate limits on expensive or risky operations
• You want a daily summary of your agent's activity

Do NOT Use This Skill When

• - You only want basic OpenClaw built-in allowlists (those work fine for simple cases)
• You don't need signed cryptographic proof (just want logs, not receipts)

Setup

Fast path (recommended)

If you already have an OpenClaw config, generate a passported wrapper pack first:

CODEBLOCK0

That writes:

• - INLINECODE1
• INLINECODE2
• INLINECODE3
• INLINECODE4
• INLINECODE5
• INLINECODE6

Copy the mcpServers entries from wrapped-config.json into your OpenClaw config.

First Run (generates signing keys + default policy)

CODEBLOCK1

This creates:

• - keys/gateway.json — Ed25519 signing keypair (in current directory)
• INLINECODE10 — default shadow-mode policy (logs everything, blocks nothing)

Wrapping Your MCP Servers

For each MCP server your agent uses, wrap it through protect-mcp. In your OpenClaw MCP config:

Before:
CODEBLOCK2

After:
CODEBLOCK3

That's it. Every tool call now produces a signed receipt.

Applying a Policy Pack

To move from shadow mode to enforce mode, copy a policy template and add --enforce:

CODEBLOCK4

Then update your MCP config args to include --enforce:
CODEBLOCK5

See the policies/ directory for pre-built templates.

Commands

When the user asks you to perform these actions, execute them:

"Show my passport" / "What's my agent identity?"

CODEBLOCK6

Display the output, which shows:

• - Total decisions (allow/deny/rate-limited)
• Time range of activity
• Top tools used
• Trust tiers seen

"What did you do today?" / "Show my daily report"

CODEBLOCK7

Display the local summary, which includes:

• - Actions taken (allowed, blocked, awaiting approval)
• Tools used and frequency
• Trust tier status
• Blocked tools with reasons

"Show my receipts" / "Prove what you did"

CODEBLOCK8

Shows the 20 most recent decisions with tool name, decision, and timestamp.

"Why was that blocked?"

When a tool call is blocked or rate-limited, explain:

• - Which policy rule triggered the block
• The agent's current trust tier
• What tier would be needed to use that tool
• How to request an approval or policy change

Approval Flow

When protect-mcp blocks a high-risk action that has require_approval: true in the policy, it returns:

CODEBLOCK9

When you receive this response:

1. 1. Tell the user: "I need your approval to use [tool_name]. Should I proceed?"
2. Wait for the user to respond with approval.
3. When the user approves, grant the approval by running:

CODEBLOCK10

Replace REQUEST_ID, TOOL_NAME, and NONCE with the values from the REQUIRES_APPROVAL response.

1. 4. After granting approval, retry the original tool call. It will now succeed.

When the user denies: Tell them the action was blocked and explain what was prevented.

Check current approvals:

CODEBLOCK11

Policy Packs

Pre-built policies are available in the policies/ directory:

Verification

Every receipt is independently verifiable. The MIT-licensed verifier requires zero trust in ScopeBlind.

Quick test (no receipts needed)

CODEBLOCK12

Getting receipts to verify

Receipts are available from the local HTTP status server while protect-mcp is running:

CODEBLOCK13

If the local status server is unavailable, use the persisted receipt file:

CODEBLOCK14

Verifying a receipt

CODEBLOCK15

The verifier checks Ed25519 signatures against public keys. No API calls, no accounts, no ScopeBlind servers involved.

Links

• - protect-mcp on npm
• @veritasacta/verify on npm
• Documentation
• GitHub