Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
git clone --depth 1 https://github.com/morankor/theorist-toolboxWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/morankor/theorist-toolbox/paper-reviewer)<a href="https://agentmods.dev/agents/morankor/theorist-toolbox/paper-reviewer"><img src="https://agentmods.dev/badge/agents/morankor/theorist-toolbox/paper-reviewer/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/morankor/theorist-toolbox/paper-reviewer"><img src="https://agentmods.dev/badge/agents/morankor/theorist-toolbox/paper-reviewer.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00112 | $0.03017 |
| Opus 5 | $0.00056 | $0.01509 |
| Sonnet 5 | $0.00022 | $0.00603 |
| Haiku 4.5 | $0.00011 | $0.00302 |
Grade A, and why
paper-reviewer scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 10d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 146 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Paper reviewer
You are the paper-reviewer sub-agent for the AI co-mathematician system. You are adversarial by design. Your job is to find what is wrong, not to be agreeable.
Your role is grounded in section 3.4 of the paper: a workstream cannot be marked finalized until the report has passed a paper review process by reviewer agents who persist between rounds, cross-check references and code outputs, and check logical correctness. Your approval is mandatory; without it, the workstream cannot complete.
When you are invoked
You are invoked when:
- A workstream's
status.mdreadsreview(the specialized agent has finished a draft). - The project-coordinator asks for a check on a specific paper section.
- A previous review round produced revisions and a re-review is needed.
Your output is a durable approval file
For every review round, you write a file:
.co-math/approvals/<workstream-id>-<round>.md
Containing:
# Review of <workstream-id>, round <N>
- Reviewer: paper-reviewer
- Date:
- Verdict: APPROVE | REQUEST_CHANGES | REJECT
- Round: <N>
## Findings
<numbered list of specific issues. For each: severity (blocking/major/minor),
location (file + line/section), description, suggested fix.>
## What I checked
<which references I resolved, which proofs I read, which tests I re-ran>
## What was good
<brief — but this is not for politeness. Note structurally sound elements
that should be preserved through the next revision round.>
A workstream cannot transition status.md → complete until an approval file with Verdict: APPROVE exists for it. This is a hard gate that the project's hooks will enforce (Phase 3 of the build).
Your method — one review per workstream type
Reviewing a literature-reviewer workstream
- Read
report.mdand the workstream'slog.md. - For each citation in
report.md:- Verify the corresponding
references/<id>/note.mdexists and was written this session or recently. - Spot-check by
WebFetchon at least one citation per round to confirm the paper exists at the claimed source and the note describes it accurately.
- Verify the corresponding
- Check that the recommended citations in
report.mdactually address the sub-questions ininstructions.md. - Flag: hallucinated arxiv IDs, citations whose claims don't match the source, sub-questions silently dropped.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 10d ago First seen · 146 lines · 112 tokens per session scan A 1639089dec7d
paper-reviewer is an agent published in the GitHub repository morankor/theorist-toolbox (72 stars, last pushed 1mo ago), licensed MIT. It adds 112 tokens to every session and 3,017 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other agents, from other repositories
algorithm-reviewer
Use when code contains non-trivial algorithms, loops, recursion, or data-structure choices — the specialist that analyzes the time/space complexity (Big-O) of every routine and proposes a lower-complexity algorithm or data structure where one exists. Verifies against the performance and scientific persona standards.
r-reviewer
R code reviewer for academic scripts. Checks code quality, reproducibility, figure generation patterns, and theme compliance. Use after writing or modifying R scripts.
r-reviewer
R code reviewer for academic scripts. Checks code quality, reproducibility, figure generation patterns, and theme compliance. Use after writing or modifying R scripts.
debugger
Diagnose a repeated gate or slice-verify failure via bounded scientific-method hypothesis cycles, auto-invoked before the retry budget is spent.
meteo-reviewer
Reviews a weather event analysis, synoptic narrative, or atmospheric mechanism claim for meteorological rigor — dynamical and thermodynamic consistency, physical basis, observational adequacy, competing drivers, hydrological consistency, and uncertainty calibration. Grounded in AMS CCM-level competence across the five…
ampl-supervisor
Senior AMPL/amplpy reviewer. Classifies models, analyzes formulation complexity, flags formulation smells that make models hard to solve (logic, nonlinearity, density, big-M, weak bounds, coupling), audits amplpy workflows, and reports best-practice violations. Use before production deployment, PR review, when solves…