# NOTICE

superred-target-inspect-agent

This package (all code under `src/inspect_agent_target/`) is original work,
Copyright (c) 2026 Simon Sure, released under the MIT License (see LICENSE).

No third-party source code or datasets are vendored or bundled inside this
package. The items below are credited for transparency; none imposes a
redistribution obligation on this wheel because none of their material is
shipped inside it.

## Wrapped framework (runtime dependency, not bundled)

Inspect (inspect_ai)
  Source:    https://github.com/UKGovernmentBEIS/inspect_ai
  License:   MIT License, Copyright (c) 2024 UK AI Security Institute
  Confirmed: https://raw.githubusercontent.com/UKGovernmentBEIS/inspect_ai/main/LICENSE
  Use:       Imported at runtime via its public API (inspect_ai.model,
             inspect_ai.tool). `run_rollout` re-implements the control flow of
             inspect's `generate(tool_calls="loop")` using inspect's own
             primitives (Model.generate + execute_tools). No inspect source is
             copied into this package; inspect-ai is pip-installed separately
             as a declared dependency.

## Configuration inspiration (no code or data used)

AgentHarm benchmark (UK AI Security Institute)
  Source:    https://github.com/UKGovernmentBEIS/inspect_evals
  License:   MIT License
  Use:       Only numeric generation defaults are mirrored as benchmark-agnostic
             constants (temperature 0.0, max_tokens 4096, seed 0, max_retries 3,
             message_limit 20). No AgentHarm code, prompts, tools, or dataset are
             included in this package.