# NOTICE

superred-target-dtap-scaffold
Copyright (c) 2026 Simon Sure

This product is licensed under the MIT License (see LICENSE).

All source code and the bundled `data/tool_trees.json` authorization mapping in
this package are original work of the superred authors. No upstream source code
or benchmark data is redistributed inside this package.

## Upstream this scaffold ports and wraps (not redistributed)

This module is a faithful superred port of, and a runtime wrapper around, the
DecodingTrust-Agent (DTap) red-teaming platform for AI agents. Its
environment-pool orchestration behavior is reimplemented from DTAP (no upstream
code is copied), and the optional `sdk` extra pulls the upstream package
`decodingtrust-agent-sdk`, which this module lazy-imports at runtime; the SDK is
NOT vendored into this wheel.

- Name:      DecodingTrust-Agent (DTap) / decodingtrust-agent-sdk
- Authors:   AI-Secure (University of Illinois / DecodingTrust team) and contributors
- Source:    https://github.com/AI-secure/DecodingTrust-Agent
- PyPI:      https://pypi.org/project/decodingtrust-agent-sdk/ (pinned ==0.2.12)
- License:   Apache-2.0 (SPDX: Apache-2.0), confirmed at
             https://github.com/AI-secure/DecodingTrust-Agent/blob/main/LICENSE

`data/tool_trees.json` references DTAP service tool names (factual API
identifiers) but is an original authorization-tree mapping authored for this
project; it is not a copy of the DTAP benchmark dataset. The DTAP task/goal
dataset is defined in the separate `security-claim-dtap` package, not here.

When reporting measurements produced with this target, cite the
DecodingTrust-Agent (DTap) platform.