# CMN-C1-637 — AWS Agent Toolkit MCP Server & Skill Governance Q&A Agent

> **Category**: Cat 1 (single technical capability, use-case-agnostic)
> **Industry**: Common (industry-agnostic)

## Overview

Question-answering over a curated knowledge base of governance guidance for a cloud provider's agent toolkit, its MCP server catalogue and agent skills. Given a question such as "What least-privilege IAM policy does an S3-read agent need?", the agent infers the scope (MCP catalogue, IAM or financial-sector data residency), retrieves matching catalogue entries, extracts a least-privilege IAM policy built from placeholders only, adds data-residency notes with region flags based on Japan's financial-sector security guidelines, and assembles a cited answer with configuration snippets. Retrieval and policy assembly are deterministic; no LLM is called. A question with no matching entry receives an explicit out-of-scope answer, an answer whose claims cannot be cited is not published, credential-shaped strings are redacted, and every answer carries an advisory disclaimer. The knowledge base shipped here is a small curated sample — replace it with your own catalogue.

This is an agent template built with the **AGENTIC STAR** development platform and the
**AgentCore Framework**. It is intended to be taken as a starting point: fork it, adapt it to
your own data and policies, and run it inside your own AGENTIC STAR deployment.

## Requirements

**This template does not run standalone.** It requires:

| Requirement | Notes |
|---|---|
| **AGENTIC STAR platform** | The agent connects to the platform at start-up. Without it, start-up fails immediately (see *Behaviour without the platform* below). Deployment guides and API documentation: [AGENTIC STAR Developers](https://developers.fd.agenticstar.tm.softbank.jp/) |
| **AgentCore Framework** (`agenticstar-agentcore`) | Installed from PyPI as a dependency. |
| Python | 3.11 or later (`requires-python = ">=3.11"`) |

```bash
pip install -e .
```

### Behaviour without the platform

The framework is designed to run **only** on AGENTIC STAR. There is no fallback or degraded
mode. If the platform is unreachable or the SDK version does not match, the agent raises
`PlatformRequired` during graph compile / start-up preflight rather than starting in a partially
working state. This is intentional — a half-running agent is worse than one that refuses to start.

## Known limitations

**Product names can be masked by the platform before retrieval.** AGENTIC STAR's personal-data
protection runs before this template's code and replaces any run of two or more Title-Case words
with `[MASKED]` — "Agent Toolkit", "Amazon Bedrock" and "Secrets Manager" included — and the
template cannot switch it off. Measured on AgentCore 1.0.3:

| Question | Reaches retrieval as | Result |
|---|---|---|
| `What does Amazon Bedrock need from IAM?` | `What does [MASKED] need from IAM?` | IAM answer without `bedrock:InvokeModel`; `limitations: ["QUERY_TERMS_MASKED"]` and a note at the top of `answer` |
| `What does amazon bedrock need from IAM?` | unchanged | includes `bedrock:InvokeModel`; `limitations: []` |
| `Explain the Amazon Bedrock setup in the AWS Agent Toolkit.` | `Explain the [MASKED] setup in the AWS [MASKED].` | `validation_status: not_evaluated` (not `out_of_scope`) |
| `How do I tune quantum annealing on Amazon Bedrock?` | `How do I tune quantum annealing on [MASKED]?` | `not_evaluated` — the unmasked words matching nothing is not treated as out of scope |
| `What does Taro Yamada recommend for ramen in Osaka?` | `What does [MASKED] recommend for ramen in Osaka?` | `not_evaluated` (an off-topic question that contains a masked name is not called out of scope either) |

When part of a question was masked, `limitations` holds the code `QUERY_TERMS_MASKED` and the
answer says so; a question with nothing left to match is `not_evaluated`, not out of scope. The
flag also appears when the masked words did not change the answer (for example "Secrets Manager"
in a question that other words already match). Write product names in lower case ("bedrock",
"agent toolkit", "secrets manager") to get a complete answer.

## Quick Start

```bash
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
python -m pytest tests/ -v
```

Tests run without a platform connection. Running the agent itself does not.

## Project Structure

```
src/          agent implementation (nodes, services, schemas)
tests/        unit, integration and boundary tests
config/       agent configuration
docs/         design and operational documentation
```

See `docs/02_design.md` for the design and `docs/03_test_spec.md` for the test specification.

## Customising

1. Adjust `config/` for your own environment and policies.
2. Replace the knowledge sources and sample data with your own.
3. Review the node implementations under `src/nodes/` for domain-specific logic.
4. Re-run the test suite.

## License

MIT — see [LICENSE](LICENSE).

## Status of this repository

This template is published **as is**, by its individual author, under the MIT license. It carries
**no warranty and no support commitment**, and no organisation stands behind its behaviour or
fitness for any purpose. Issues and pull requests may or may not receive a response; that is at
the sole discretion of the repository owner.