agentest Assurance Report Card

agentest Assurance Report Card

Checks an AI agent's prompt or recorded trace against an open test checklist and returns a report card, offline in your browser. For agent builders and testers who want repeatable checks.

Made by srinivaspadmanabhuni on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: This Space gives an AI agent a report card. You paste either the agent's prompt or logic, or a recorded trace of a run, and it checks the text against an open Agentic AI Test Checklist. The checks run as Python inside your browser, so no model is called and nothing leaves your machine.

Who it suits: Agent developers, testers and reviewers who want to find gaps in an agent's controls before it goes live.

Industry | Who uses it | How they use it | Example
Software | Agent developers | Review an agent prompt for missing controls | Paste a prompt that handles payments and read the gaps
Quality assurance | Testers | Judge a recorded run against numbered checks | Paste a trace and read the pass, fail and n/a verdicts
Finance | Risk reviewers | See whether money moving steps state their limits | Check an agent that can move funds
Security | Security reviewers | Spot instructions that work against a control | Look for risky instructions in a prompt

If you want a tool that runs your agent for you, this page reads text and traces and does not run anything.

Where to use it: It needs only a browser.

• A web browser, on the Space's page at Hugging Face with nothing to install
• A downloaded copy on Windows, macOS or Linux, opened in a browser
• A separate command line project, for CI and pytest use

Use case | Who | What they get back
Review a prompt | Developers | For each risk, Stated, Gap or Risky instruction, with a checklist ID
Judge a trace | Testers | Pass, fail or n/a for each checklist ID, with evidence
Gate a build | CI owners | Reports in JUnit, SARIF, JSON, HTML, Markdown or CSV

How to set it up: Using it online needs nothing. A local copy of the page needs two things.

• Python and pip
• A web browser

1. Install the Hugging Face tools.
2. Download the Space into a folder.
3. Move into that folder.
4. Open index.html in your browser.

pip install -U huggingface_hub
hf download srinivaspadmanabhuni/agentest-assurance-report-card --repo-type space --local-dir agentest-assurance-report-card
cd agentest-assurance-report-card

How to use it: Open the page and pick a tab. On the prompt tab, paste the agent's prompt or logic. On the trace tab, paste an OTLP or JSON trace. Read the report card that comes back. Do not paste customer data or credentials. The command line project can also receive traces and import framework logs.

agentest std serve
agentest std import

Tool | What it does
Prompt or logic tab | A rule based design review that reads the text only
Trace tab | Judges a pasted trace and reports pass, fail or n/a per ID
agentest std serve | Receives traces over OTLP and HTTP
agentest std import | Converts native framework logs into traces

Example: You paste the prompt of an agent that can send messages and handle personal data. The report card marks which risks the prompt covers, which it leaves out, and which instructions work against a control, each with a checklist ID and a hint for testing on a trace.

For newcomers and experts: Start on the prompt tab with a short prompt and read the green, red and orange verdicts. Each one names a checklist ID you can look up.

Judging is deterministic. No language model sits in the pass or fail path, so the same trace gives the same verdict each time. Rules read standard evidence such as tool annotations, call arguments, results and span status, and they never name a vendor. A check the trace holds no evidence for reads n/a and never counts as a quiet pass.

You paste a prompt or a trace
Python runs in the browser through Pyodide
Rules map the text to checklist IDs
The report card shows each verdict and its detail

Good to know: The page does not run your agent, and a trace is judged only after the fact.

Title: Checks by origin
Checklist checks: 110 checks
Repo extensions: 20 checks

• The command line project needs only PyYAML, with no API key, GPU or model
• The Python package is not on PyPI, so it comes from the git repository
• License: Apache 2.0
• Cost to run: none, since no account, key, network or model is needed
• Data it reaches: only what you paste, which stays in your browser
• Not for: running the agent itself or pasting customer data, credentials or blind round material

On Hugging Face, 13 November 2026

• Likes: 1
• Built with: static
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/srinivaspadmanabhuni/agentest-assurance-report-card
• Files: https://huggingface.co/spaces/srinivaspadmanabhuni/agentest-assurance-report-card/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.