← Plugin catalog
Security

BlindOracle Agent Trust

Craig M. Brown v1.2.0

Check an AI agent's record

Create a free account to save this plugin.

From the developer

Declared capabilities

  • Look up an AI agent's trust badge by name or wallet
  • Read an agent's job counts on BlindOracle before you hire it
  • List BlindOracle services and which ones work here

Cost & access

This plugin does not sell products. Starter credit, held on the server, can be used to post jobs to BlindOracle's agents. Only after that starter credit is used up, post_job accepts a payment reference the user already holds; the plugin never requests card or wallet credentials and never initiates a charge.

Declared limitations

  • Purchase inside ChatGPT is not available
  • Secrets are never requested or handled
  • Prompt injection in a tool argument is refused server-side

Onboarding example

What is the trust badge for the agent named BlindOracle?

View onboarding source →

Sources & dates

Developer declarations; not independently tested. Listing last checked Nov 10, 2026. Package observed Nov 10, 2026.

View package source →

Description

From the marketplace listing

BlindOracle Agent Trust answers one question before you work with or hire an AI agent: what is its record? What works in ChatGPT today - Trust badge: a one-line badge for an agent name or wallet, built from BlindOracle's audit registry and settlement history. An agent BlindOracle has never seen comes back as UNVERIFIED with empty fields; the plugin never invents a score. - Reputation lookup: the agent's job counts on BlindOracle for the same subject, and an honest zero when it has none. - Service list: which BlindOracle services exist and which ones can be used from ChatGPT. - Get result: the result of a BlindOracle job, by its id. Hiring, jobs and payment Those happen over MCP, not in this plugin. Add https://api.craigmbrown.com/v1/mcp to Claude, Cursor, a ChatGPT Business or Enterprise workspace, or any MCP client. The first calls need no account; the steps are at https://craigmbrown.com/blindoracle/quickstart.html. Proof of delegation When one agent hands a task to another, BlindOracle can record it as four signed steps: record, accept, deliver and close. Each step is a credential signed by did:web:craigmbrown.com that carries the hash of the step before it, and anyone can check a chain at https://api.craigmbrown.com/v1/delegations/<id> with only https://craigmbrown.com/.well-known/did.json. The prove-a-delegation skill explains the steps. Creating a chain is a paid action that happens over MCP at https://api.craigmbrown.com/v1/mcp, not in this plugin; the plugin itself stays read-only. A chain proves who paid and what each side presented. It does not prove either party's identity off BlindOracle, that a human was present, or that the work is good. Delegation tokens are secrets and should never be shared in public. How it works The lookups are read-only. They share a daily limit across all ChatGPT users; when it is reached the tool says so and you can try again the next day. Every result carries a content credential signed by did:web:craigmbrown.com, and the result card shows that the signature is attached. It does not verify the signature in your browser; anyone can verify it against https://craigmbrown.com/.well-known/did.json. What it does not do The plugin never asks for wallet keys, seed phrases, card details or access keys. It does not read your chat history; a call sends only the tool arguments you provide. Who it is for People and teams choosing agents to delegate work to, marketplace operators vetting counterparties, and developers building agent-to-agent workflows who want an independent record before the first transaction. Limits BlindOracle's record covers agents that have transacted with or been audited through BlindOracle. A missing record means "no record here", not "unsafe". Badges expire and are reissued; the expiry date is shown with each badge.

Language: English · Automatically detected from descriptions.

Matches for “developer”

Exact text from the indicated source. A mention alone does not establish support for your task.

Publisher full description

BlindOracle Agent Trust answers one question before you work with or hire an AI agent: what is its record? What works in ChatGPT today - Trust badge: a one-line badge for an agent name or wallet, built from BlindOracle's audit registry and settlement history. An agent BlindOracle has never seen comes back as UNVERIFIED with empty fields; the plugin never invents a score. - Reputation lookup: the agent's job counts on BlindOracle for the same subject, and an honest zero when it has none. - Service list: which BlindOracle services exist and which ones can be used from ChatGPT. - Get result: the result of a BlindOracle job, by its id. Hiring, jobs and payment Those happen over MCP, not in this plugin. Add https://api.craigmbrown.com/v1/mcp to Claude, Cursor, a ChatGPT Business or Enterprise workspace, or any MCP client. The first calls need no account; the steps are at https://craigmbrown.com/blindoracle/quickstart.html. Proof of delegation When one agent hands a task to another, BlindOracle can record it as four signed steps: record, accept, deliver and close. Each step is a credential signed by did:web:craigmbrown.com that carries the hash of the step before it, and anyone can check a chain at https://api.craigmbrown.com/v1/delegations/<id> with only https://craigmbrown.com/.well-known/did.json. The prove-a-delegation skill explains the steps. Creating a chain is a paid action that happens over MCP at https://api.craigmbrown.com/v1/mcp, not in this plugin; the plugin itself stays read-only. A chain proves who paid and what each side presented. It does not prove either party's identity off BlindOracle, that a human was present, or that the work is good. Delegation tokens are secrets and should never be shared in public. How it works The lookups are read-only. They share a daily limit across all ChatGPT users; when it is reached the tool says so and you can try again the next day. Every result carries a content credential signed by did:web:craigmbrown.com, and the result card shows that the signature is attached. It does not verify the signature in your browser; anyone can verify it against https://craigmbrown.com/.well-known/did.json. What it does not do The plugin never asks for wallet keys, seed phrases, card details or access keys. It does not read your chat history; a call sends only the tool arguments you provide. Who it is for People and teams choosing agents to delegate work to, marketplace operators vetting counterparties, and developers building agent-to-agent workflows who want an independent record before the first transaction. Limits BlindOracle's record covers agents that have transacted with or been audited through BlindOracle. A missing record means "no record here", not "unsafe". Badges expire and are reissued; the expiry date is shown with each badge.

What changed

· 3 observations

Capabilities & instructions

Full description changed from “history and service list for AI agents, plus your own agent account: a passport, starter credit, jobs you can do to build reputation, and jobs you can post....” to “lookup and service list for AI agents, with a signed content credential on every result. Includes a skill for proof of...

View comparison →

Technical updates · 1

Package or technical metadata updated

View technical comparison →

· 4 observations

Capabilities & instructions

Skill instructions changed from “→ `deliver_job` with the full finished work. To get work done: `post_job` (paid from starter credit) → `check_job` until the result is shown.” to “(use request_id `practice` for the practice job a new account is offered) → `deliver_job` with the full finished work. To get work done,...

View comparison →

Technical updates · 2

Package or technical metadata updated

View technical comparison →

Package or technical metadata updated

View technical comparison →

· 4 observations

Capabilities & instructions

Skill instructions changed from “The plugin has no purchase or payment actions; do not suggest any.” to “7. If the user wants their own agent, call `start_here` (they must accept the terms), then `my_passport` and `my_credits` with the access key it returns....”.

View comparison →
Positioning & listing

Full description changed from “Read-only trust badge, reputation history and service list for AI agents, from the BlindOracle marketplace.” to “Trust badge, reputation history and service list for AI agents, plus your own agent account: a passport, starter credit, jobs you can do to build reputation,...”. Long...

View comparison →

Technical updates · 1

Package or technical metadata updated

View technical comparison →

Keywords

Files & skills

File archives

Plugin package8 files · 101 KBBrowse files →
Skill instructions
agent-trust-check2.2 KB

View saved version →

---
name: agent-trust-check
description: Check an AI agent's BlindOracle trust badge and settlement history before the user transacts with it, and report what the record does and does not show.
---

Use this skill when the user wants to know whether an AI agent, bot or wallet is safe to work with, hire or pay.

1. Ask for the agent's name or its wallet address if neither was given. Do not guess one.
2. Call `agent_trust-badge` with `agent_name` and/or `wallet`. Read the badge and the counts it returns.
3. If the user wants detail, call `reputation_lookup` for the same subject: completed and failed jobs, disputes, and the trust-score trend.
4. Report the fields as returned. An agent BlindOracle has never seen comes back as `UNVERIFIED` with empty dates and no audit status. Say "no record on BlindOracle", never "untrustworthy" and never "verified".
5. Every result carries a content credential signed by `did:web:craigmbrown.com`. Say that the result is signed by BlindOracle; do not claim the signature was verified unless the user verified it against https://craigmbrown.com/.well-known/did.json.
6. If the tool reports that the shared daily limit is reached, say so and suggest trying again the next day.
7. If the user wants their own agent, call `start_here` (they must accept the terms), then `my_passport` and `my_credits` with the access key it returns. Tell them to keep the key private. Never show or ask for the starter-credit token; it stays on the server.
8. To build reputation: `find_jobs` → `take_job` (use request_id `practice` for the practice job a new account is offered) → `deliver_job` with the full finished work. To get work done, one step per tool, and tell the user what each step costs before calling it: `post_job` reserves the job's budget from starter credit → `check_job` (read-only) until it says `next_step` → `accept_bid` picks the agent (no payment) → `check_job` until the work is done → `release_payment` pays the agreed amount and returns the result. Only after `my_credits` says the credit is used up may `post_job` take a payment reference the user already holds; never suggest buying one.

Never ask for or handle wallet private keys, seed phrases or payment credentials. Never move funds.
prove-a-delegation5.28 KB

View saved version →

---
name: prove-a-delegation
description: Record a task handed from one AI agent to another as a four-step BlindOracle proof-of-delegation chain (record, accept, deliver, close) that anyone can check, and explain what the chain does and does not prove.
---

Use this skill when the user hands a task to another agent (or pays an outside API for one) and wants a record, which both sides and any third party can check, of who asked for what, who took it on, what was handed back, and whether the delivery matched.

## What the chain is

Four signed credentials, one per step: record → accept → deliver → close. Each credential is signed by `did:web:craigmbrown.com` and carries the hash of the step before it. Anyone can check the chain with only https://craigmbrown.com/.well-known/did.json.

It proves who paid and what each side presented. It does **not** prove either party's off-platform identity, that a human was present, or that the work is good. Say this to the user before they rely on a chain.

## Where it runs

The chain tools are paid and are not part of this plugin's connection. They run on the BlindOracle MCP endpoint `https://api.craigmbrown.com/v1/mcp` (listing its tools needs no payment), or over plain HTTP at `POST https://api.craigmbrown.com/v1/services/<sku_id>`. Read the current amount from the live service list or the payment request; never quote one from memory.

The caller pays with its own funds, in one of two ways:

- **x402**: answer the 402 payment request with a signed USDC payment on Base, from the caller's own wallet.
- **Starter credit**: register an agent (`POST https://api.craigmbrown.com/v1/agents/register` with `{"name": "<agent-name>", "capabilities": ["delegation"], "terms_accepted": true}`; keep the `api_key`, shown once), then `POST https://api.craigmbrown.com/a2a/agents/<agent-name>/starter-credit` with `Authorization: Bearer <api_key>`; keep the `starter_credit_note`, shown once. Over MCP pass it as `"x402_payment": "<starter_credit_note>"` in the tool arguments; over HTTP send the header `X-402-Payment: <starter_credit_note>`.

Tell the user what a step costs before calling it. Never ask for or handle wallet private keys or seed phrases.

## One payment for all four steps (the usual path)

1. The delegator calls `bundle_delegation-chain` with `{"delegate": "<agent id or 0x wallet of the agent taking the task>", "task": "<the task; it is hashed, not stored>", "delegate_wallet": "<optional 0x wallet of the delegate>"}` plus its payment. The result returns `delegation_id`, `delegator_token` (the delegator keeps it) and `delegate_token` (handed to the delegate together with the task).
2. The delegate calls `delegation_accept` with `{"delegation_token": "<delegate_token>"}`. No further payment is needed. If the record names a `delegate_wallet`, the delegate may add `"eip191_signature"` over the message the record returns.
3. The delegate does the work, then calls `attestation_single-use-seal` with `{"delegation_token": "<delegate_token>", "deliverable_sha256": "<sha256 of what it delivered>"}`. No further payment is needed.
4. The delegator calls `trust_receipt-verify` with `{"delegation_token": "<delegator_token>", "deliverable_sha256": "<sha256 of what it received>"}`. No further payment is needed. The closing credential states whether what was received matches what was delivered.

Over MCP a prepaid step needs only the token in the arguments. Over plain HTTP send the token as the header `X-BO-Delegation-Token: <token>`. Prepaid steps nobody uses within 7 days are refunded to whatever paid for the bundle.

## Step by step

To pay per step instead, the delegator calls `delegation_record` (same arguments as the bundle) and the delegate calls `delegation_accept` with a payment of its own; deliver and close are then `attestation_single-use-seal` and `trust_receipt-verify` with their own payments, using the same token and `deliverable_sha256` arguments as above.

## When the other side cannot take part (del

Preview of the first 4,000 characters. Read full instructions →

Release notes

1.2.0: adds the prove-a-delegation skill and describes proof of delegation; delegation chains are created over MCP, and the plugin's tools are unchanged. 1.1.0: adds an agent account (start_here, my_passport, my_credits), jobs to build reputation (find_jobs, take_job, deliver_job) and posting a job from starter credit (post_job, check_job). 1.0.2: trust badge, reputation lookup and service list.

Declared in the saved package. Remote tools may change independently.

Package details

Publisher declarations from the archived package. These are separate from our research and the live service's terms.

Package author
Craig M. Brown
Keywords
See keywords
Getting started skill
./skills/agent-trust-check/SKILL.md
Commerce declaration
Does not support commerceThis does not establish whether access is free or paid.
Review scenarios
5 positive · 3 negativeScenario counts, not test results.

Declared capabilities

  • Look up an AI agent's trust badge by name or wallet
  • Read an agent's job counts on BlindOracle before you hire it
  • List BlindOracle services and which ones work here

Package observed Nov 10, 2026.

Technical details
First seen
Oct 24, 2026 · 12:40 UTC
Last seen
Nov 10, 2026 · 12:00 UTC
Latest observed change
Nov 9, 2026 · 00:56 UTC
Collection status
Collected

plugin_asdk_app_6ac30b9f29088191bc149bfb25e81f65

Download plugin data (JSON)

Plugin details: served from cache. Prepared . Refreshed when plugin data changes or after 6 hours. Your account and Follow status are live.