---
title: "Math proof worker, deep: review, role & definition · Undominated.ai"
canonical: https://undominated.ai/agents/anthropic-math-proof-worker-deep/
description: "Pins the higher-effort math-proof worker that answers one question per launch and writes the answer to a file the orchestrator reads."
---

# Math proof worker, deep: review, role & definition · Undominated.ai

> Pins the higher-effort math-proof worker that answers one question per launch and writes the answer to a file the orchestrator reads.

[← Explore all agents](/agents/)

MATHEMATICS / Anthropic

# Math proof worker, deep

Pins the higher-effort math-proof worker that answers one question per launch and writes the answer to a file the orchestrator reads.

 [Use this definition ↓](#setup)[Original source ↗](https://github.com/anthropics/claude-plugins-official/tree/41f32e407b450055044f167db7ded5e8ff67d203/plugins/math-proof/agents)

SOURCE REVIEW

 Reviewed 2026-10-14
 Evidence 3 linked sources
 Publisher Anthropic
 Licence [Apache-2.0 ↗](https://github.com/anthropics/claude-plugins-official/blob/41f32e407b450055044f167db7ded5e8ff67d203/plugins/math-proof/LICENSE)
 Revision 41f32e407b45
 [Read what was—and wasn’t—checked ↓](#review)

“A siege-skill worker with Read, Write, and Edit only, no memory between questions, and a reply capped as a short abstract of the answer file.”

 [Anthropic · upstream description ↗](https://github.com/anthropics/claude-plugins-official/blob/41f32e407b450055044f167db7ded5e8ff67d203/plugins/math-proof/agents/math-proof-worker-deep.md) Our analysis follows below.

01 / THE REASONING

## Why this made the selection.

 - The catalogue has math-proof-worker and math-proof-judge and not this deep worker.
- The file states that it is launched only by the siege skill, that it has no memory between questions, and that its tools are Read, Write, and Edit.

### A good fit for

 - A single self-contained proof question inside the math-proof plugin's later siege round
- Provide one query file, authoritative problem statement and designated answer file.
- Receive a checkpointed mathematical argument with proved, sketched and open parts separated.

### Weigh up before choosing

 - This review did not execute the resource.
- The file says it starts at round 4 unless ESC_ROUND says otherwise. That setting was not read from the siege skill.
- maxTurns 80 and effort medium are frontmatter values, not a measurement of a run.
- Read/Write/Edit permit answer artifacts but provide no shell, formal prover or web retrieval; cited literature is memory-derived.

02 / THE REVIEW RECORD

## What we actually inspected.

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

### Material inspected

 - plugins/math-proof/agents/math-proof-worker-deep.md
- plugins/math-proof/LICENSE
- README.md
- Round 2: pinned definition, output contract and permission/dependency boundaries reread on 2026-10-14

### Our findings

 - The description says the worker has no memory between questions and is launched only by the math-proof plugin's siege skill.
- Frontmatter tools are Read, Write, and Edit, with model inherit, effort medium, and maxTurns 80. Bash is not listed.
- The body says the answer file must separate what is proved, sketched, or open, and that the orchestrator sees the short reply directly while the file is what gets read closely.
- The worker may reuse a named earlier partial answer only after independently checking it, not as established proof.
- The final sentinel marks a finished attempt even when the task remains unresolved; it is not a theorem-validity signal.

### Not established by this review

 - The resource was not executed and its tools were not called.
- The siege skill was not opened.
- No proof task was given to the worker.

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 ↗](https://github.com/anthropics/claude-plugins-official/blob/41f32e407b450055044f167db7ded5e8ff67d203/README.md)
 - The README's install line is /plugin install {plugin-name}@claude-plugins-official and does not name math-proof.
- install.command is null because that placeholder is not a finished command.
- This agent is plugins/math-proof/agents/math-proof-worker-deep.md. The plugin LICENSE is Apache-2.0.
- Dispatch through the matching siege workflow, preserve the exact sentinel and keep independent proof checking separate from completion detection.

### Before you start

 - Claude Code
- The math-proof plugin's siege skill, which this review did not open

### Compatibility

Claude Code with the math-proof plugin

### Single-question worker that writes an answer file

 - Read files the prompt names
- Write and edit the answer file
- No shell tool is declared

### Cost model

Apache-2.0; the worker was not launched

04 / FOLLOW THE EVIDENCE

## The source trail.

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

 - [math-proof-worker-deep ↗](https://github.com/anthropics/claude-plugins-official/blob/41f32e407b450055044f167db7ded5e8ff67d203/plugins/math-proof/agents/math-proof-worker-deep.md) Checked 2026-10-14 https://github.com/anthropics/claude-plugins-official/blob/41f32e407b450055044f167db7ded5e8ff67d203/plugins/math-proof/agents/math-proof-worker-deep.md Supports: tool list, no memory, and answer-file rule
- [math-proof LICENSE ↗](https://github.com/anthropics/claude-plugins-official/blob/41f32e407b450055044f167db7ded5e8ff67d203/plugins/math-proof/LICENSE) Checked 2026-10-14 https://github.com/anthropics/claude-plugins-official/blob/41f32e407b450055044f167db7ded5e8ff67d203/plugins/math-proof/LICENSE Supports: Apache License Version 2.0
- [Repository README ↗](https://github.com/anthropics/claude-plugins-official/blob/41f32e407b450055044f167db7ded5e8ff67d203/README.md) Checked 2026-10-14 https://github.com/anthropics/claude-plugins-official/blob/41f32e407b450055044f167db7ded5e8ff67d203/README.md Supports: placeholder install line

KEEP COMPARING

## Other approaches to consider.

Related by category or shared topics. These are alternatives to inspect, not a measured quality order.

 [### Math-proof judge ↗ Runs one named step of a math-proof siege: writes only the files that step names, keeps notes under 4000 words, and requires proof.md to stand alone.](/agents/anthropic-math-proof-judge/)[### Math-proof worker ↗ Answers one self-contained mathematics question into a file, separates proved claims from sketches and open questions, and appends the task's end marker only when finished.](/agents/anthropic-math-proof-worker/)[### C review worker ↗ Pins the c-review producing worker that reads source, writes one part file, and states that it has no shell.](/agents/trailofbits-c-review-worker/)

 [AI Tools ↗](/tools/)[Skills ↗](/skills/)[Agents ↗](/agents/)[MCP Servers ↗](/mcp-servers/)[Workflows ↗](/workflows/)

## Continue your investigation

 - [Find reusable instructions](/skills/)
- [Inspect connections](/mcp-servers/)
- [Build a toolkit](/tools/)
- [Choose the model](/compare/)
