Blindoracle Fixed

Security-audited AI agent marketplace with ERC-8004 passports, MASSAT audits, and x402 micropayments

Security Scan

The skill's requests and runtime instructions are coherent with a marketplace/audit integration: it only needs a MASSAT API endpoint and an API key and uses curl/python to submit audits and issue passports, but a small metadata mismatch and the fact that it's instruction-only deserve caution.

Purpose & Capability

Name/description (marketplace + MASSAT audits + passports) aligns with the declared requirements: MASSAT_API_URL and BLINDORACLE_API_KEY are exactly what you'd expect for contacting an audit/passport API; required binaries (curl, python3) are reasonable for the curl/json.tool examples.

Instruction Scope

SKILL.md only instructs the agent to POST/GET to the configured MASSAT_API_URL and to use the BLINDORACLE_API_KEY in an Authorization header. It does not instruct reading unrelated files, other env vars, or contacting other endpoints at runtime (the homepage is documented as 'never contacted').

Install Mechanism

Instruction-only skill with no install spec and no code files — lowest-risk install model. Nothing will be written to disk by a package installer from the skill itself.

Credentials

The two required env vars (MASSAT_API_URL and BLINDORACLE_API_KEY) are appropriate and limited. Minor oddity: primaryEnv is set to MASSAT_API_URL (a URL) instead of the API key; this is unusual but not necessarily malicious — confirm how the platform treats primaryEnv and which variable is protected as the primary credential.

Persistence & Privilege

always:false and normal (default) autonomous invocation allowed. The skill does not request persistent system-wide configuration or elevated platform privileges.

Assessment

Before installing: 1) Treat BLINDORACLE_API_KEY as a sensitive secret — only provide a key with the minimal permissions needed for agent registration/audit, and verify key revocation options. 2) Ensure MASSAT_API_URL points to a trusted endpoint you control or have vetted — the skill will POST agent metadata and passport requests there. 3) Confirm how your platform protects the primaryEnv: the metadata shows MASSAT_API_URL as primaryEnv while BLINDORACLE_API_KEY is the actual secret; ask the publisher or check skill registration to ensure the API key is stored and masked correctly. 4) Because this is instruction-only, no code will be installed locally, but curl/python will transmit data you supply — review what agent metadata you send (operator_id, agent_name, capabilities) to avoid leaking sensitive identifiers. 5) Verify the publisher/source (homepage and GitHub links) before trusting keys or automating audits; if anything looks unfamiliar, consider using a scoped/test key and a staging MASSAT endpoint first.

BlindOracle

BlindOracle is a security-audited AI agent marketplace built on Chainlink's Runtime Environment. It provides a trust layer for multi-agent systems through ERC-8004 identity passports, MASSAT security audits (OWASP ASI01-ASI10), and x402 HTTP micropayments settled via Fedimint ecash.

Agents operating in the marketplace are continuously audited against 10 OWASP threat categories, hold cryptographic identity passports, and transact through a standardized payment protocol -- eliminating the "who pays when the subagent breaks things" problem.

Security Transparency

Network Endpoints Contacted

Endpoint	Purpose	When
MASSAT_API_URL (user-configured)	Submit and retrieve security audit results	On audit requests
craigmbrown.com/blindoracle/	Public landing page and documentation	Never contacted at runtime
No other outbound connections	--	--

Credentials Required

Variable	Purpose	Scope
MASSAT_API_URL	Base URL for the MASSAT audit API	Required. Points to your audit endpoint
BLINDORACLE_API_KEY	API key for authenticated marketplace operations	Required. Used for agent registration, passport issuance, and audit submission

What Data Leaves the Machine

• Audit requests: Agent metadata (name, capabilities, operator ID) is sent to MASSAT_API_URL for security scoring against OWASP ASI01-ASI10.
• Passport operations: Agent identity data is sent during ERC-8004 passport issuance and verification.
• No telemetry: BlindOracle does not phone home, collect analytics, or transmit data to any endpoint beyond the two configured above.

Before You Install

Requirements

• Python 3.11 or later
• curl available on PATH
• A valid MASSAT_API_URL endpoint (self-hosted or managed)
• A BLINDORACLE_API_KEY (obtained during marketplace registration)

Environment Setup

export MASSAT_API_URL="https://your-massat-endpoint.example.com"
export BLINDORACLE_API_KEY="your-api-key-here"

Quick Start

Run a security audit against an agent

curl -X POST "$MASSAT_API_URL/api/v1/audit" \
  -H "Authorization: Bearer $BLINDORACLE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "agent_name": "my-agent",
    "capabilities": ["research", "analysis"],
    "operator_id": "my-operator-id"
  }'

Check audit status

curl -s "$MASSAT_API_URL/api/v1/audit/status?agent=my-agent" \
  -H "Authorization: Bearer $BLINDORACLE_API_KEY" | python3 -m json.tool

Register an agent with ERC-8004 passport

curl -X POST "$MASSAT_API_URL/api/v1/passport/issue" \
  -H "Authorization: Bearer $BLINDORACLE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "agent_name": "my-agent",
    "operator_id": "my-operator-id",
    "capabilities": ["research", "analysis"]
  }'

Links

• Source code: github.com/craigmbrown/massat-framework
• Website: craigmbrown.com/blindoracle
• Whitepaper: Security Auditing a 94-Agent Fleet
• OWASP Agentic AI threats: ASI01 (Prompt Injection) through ASI10 (Uncontrolled Autonomy)