← Explore all agents

EVIDENCE REVIEW / Undominated.ai

Undominated · Safety Refusal Calibrator

Calibrates safety classifier thresholds to minimize benign query refusals while maintaining strict compliance boundaries.

“Calibrates safety classifier thresholds to minimize benign query refusals while maintaining strict compliance boundaries.”

01 / THE REASONING

Why this made the selection.

  • Defines a specialist role, required evidence, handoff contract and explicit limits on external actions.
  • Created for recurring evidence and comparison failures: missing facts remain unknown and every conclusion keeps its scope.
  • It names a concrete failure mode: Deploy when developer teams experience over-refusal friction on technical or domain-specific workflows.

02 / THE REVIEW RECORD

What we actually inspected.

Source review has boundaries.
A clear record is more useful than a “safe” badge.

Material inspected

  • agents/undominated-safety-refusal-calibrator/AGENT.md

Our findings

  • First-party resource authored by Undominated.ai; this is our own verification record, not an independent endorsement.
  • The download preserves the complete source and supporting files. Its manifest hashes identify the exact bytes.
  • One input rule the profile states is: Review false-refusal logs, evaluate classifier confidence margins, and tune system prompt disambiguation instructions.
  • One refusal the profile states is: Does not loosen boundaries on core safety violations (CSAM, CBRN, self-harm). Calibrates benign edge cases.

Not established by this review

  • Model compliance with these instructions on arbitrary tasks
  • Native integration in every agent host
  • Public GitHub source identity
  • The new npm resource CLI installation

The review applies to the material and revision named here. A newer upstream release can change its behavior.

03 / PUT IT TO WORK

Use the role in your project.

Upstream setup instructions ↗
  1. Download the complete bundle from https://undominated.ai/resources/agents/undominated-safety-refusal-calibrator/bundle.zip and extract it into a new directory.
  2. Read AGENT.md, then load or adapt it using your host’s documented role or subagent mechanism.
  3. Provide the task scope, accepted evidence and required handoff fields. Grant only the tools the task needs.

Before you start

  • An agent host that can accept explicit role instructions

THE COMPLETE REVIEWED DEFINITION

Read it before you reuse it.

Original source bytes, with attribution.
Review the host-specific setup notes above.

---
name: undominated-safety-refusal-calibrator
description: Calibrates safety classifier thresholds to minimize benign query refusals while maintaining strict compliance boundaries.
license: MIT
metadata:
  author: Undominated.ai
  version: "1.0.0"
  tags: ["safety", "refusal", "calibration", "compliance", "developer-experience"]
---

# Safety Refusal Calibrator

## Mission & Boundary
Calibrates safety classifier thresholds to minimize benign query refusals while maintaining strict compliance boundaries.

## When to Deploy
Deploy when developer teams experience over-refusal friction on technical or domain-specific workflows.

## Operational Boundaries & Limitations
Does not loosen boundaries on core safety violations (CSAM, CBRN, self-harm). Calibrates benign edge cases.

## Execution Protocol
Review false-refusal logs, evaluate classifier confidence margins, and tune system prompt disambiguation instructions.

The download contains AGENT.md. Keep its filename when placing it in the agent directory described above.

By Undominated.ai. Exact upstream source ↗ · Licence · Attribution

SHA-256 4ada799376245349b46dcb09086f7827280a0a84aff19edf3a6b57846393af0e

Read the applicable licence
MIT License

Copyright (c) 2026 Undominated.ai

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

04 / FOLLOW THE EVIDENCE

The source trail.

Our notes are separate from the original resource.
Check upstream before adopting a new version.

  1. https://undominated.ai/resources/agents/undominated-safety-refusal-calibrator/definition.md

    Supports: Workflow and input requirements, Stated limitations

  2. MIT licence ↗Checked

    https://undominated.ai/resources/agents/undominated-safety-refusal-calibrator/LICENSE.txt

    Supports: Redistribution terms and attribution

  3. Complete source bundle ↗Checked

    https://undominated.ai/resources/agents/undominated-safety-refusal-calibrator/bundle.zip

    Supports: Full local source, supporting files and integrity manifest

Evidence & Ask