offload-policy

offload-policy is a cursor rule for Cursor from Sounio-lang/sounio. It costs 739 tokens per session, scanned A, original, Apache-2.0.

A repository rule that requires outside AI reviews at set checkpoints for mathematical claims, clinical code, and papers.

In plain words
What is it for?
It guides agents on when to run math, clinical, or document reviews, which review tools to use, and how to record findings.
Why use it?
It reduces the risk of unnoticed errors in high-stakes technical and research work by making reviews and their records mandatory.

Cursor rule for Cursor

Install

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.

agentmods
npx agentmods add rules/sounio-lang/sounio/offload-policy
Clone the repo
git clone --depth 1 https://github.com/Sounio-lang/sounio

Made for: Cursor.

Wrote 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.

agentmods badge for offload-policy

README.md
[![agentmods](https://agentmods.dev/badge/rules/sounio-lang/sounio/offload-policy.svg)](https://agentmods.dev/rules/sounio-lang/sounio/offload-policy)
Your own site
<a href="https://agentmods.dev/rules/sounio-lang/sounio/offload-policy"><img src="https://agentmods.dev/badge/rules/sounio-lang/sounio/offload-policy.svg" alt="Measured on agentmods" height="20"></a>
Per session 739 This file is loaded in full into every session.
When invoked 739 The same file — it is already loaded in full.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00739 $0.00739
Opus 5 $0.00369 $0.00369
Sonnet 5 $0.00148 $0.00148
Haiku 4.5 $0.00074 $0.00074

Measured 3d ago against content hash 78865b5daff8, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

offload-policy 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 3d 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.

.cursor/rules/offload-policy.mdc · 44 lines

How it starts

The opening of the file, as written. The whole thing — 44 lines — stays where its author put it; the contents beside it link to each section on GitHub.

LLM-Offload Policy (MANDATORY)

Canonical doc: .claude/AGENT_OFFLOAD_POLICY.md. Routing: .claude/offload-routing.md. Audit log: .claude/llm_offload_log.md.

This rule applies to every Cursor agent session in this repo, including subagents and best-of-N runners.

Mandatory pre-commit checkpoints

Trigger Required action
Math claim in .sio comments, PK/GUM/p-box derivation, Lean theorem statement, refinement invariant bin/llm-offload -t math-review -p xai -i <file>
Any change under stdlib/clinical/, tests/run-pass/vancomycin*, tests/stdlib/clinical/, formal/lean4/SounioVancomycin* bin/llm-offload -t review -p deepseek -i <file>
Any external-facing artifact under docs/papers/, docs/dissertation/, IRB protocols bin/llm-offload --raw <draft> deepseek xai gemini

After every non-trivial offload, append a row to .claude/llm_offload_log.md (top of "Entries" table). When a review caught a bug, add an LLM-offload-review: trailer to the commit message.

Quick reference

bin/llm-offload --status                        # which API keys are loaded
bin/llm-offload --list-tasks                    # available tasks + default providers
bin/llm-offload --list-providers                # all providers
bin/llm-offload -t <task> -p <provider> -i <file>
bin/llm-offload --raw <prompt> [providers...]   # multi-model fan-out

Tasks: expand (Gemini, prose), scaffold (DeepSeek, boilerplate), review (DeepSeek, devil's advocate), paraphrase (MiniMax/Qwen, tone shift), math-review (Grok 4.1, math audit).

Hard rules

  • NEVER send PHI / patient-identifying data to any provider.
  • NEVER silently dismiss a reviewer finding; always log disagreement with reasoning.
  • NEVER commit math-touching changes to stdlib/clinical/ or formal/lean4/ without a logged math-review evidence.
  • NEVER edit .claude/AGENT_OFFLOAD_POLICY.md, .claude/offload-routing.md, or bin/llm-offload in parallel with other agents (high-risk shared control files).

Read the full file on GitHub · 44 lines

Changes

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.

  1. 3d ago First seen · 44 lines · 739 tokens per session scan A 78865b5daff8

Subscribe to this mod's changes

offload-policy is a cursor rule published in the GitHub repository Sounio-lang/sounio (5 stars, last pushed 3d ago), licensed Apache-2.0. It adds 739 tokens to every session, about $0.0037 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 2026-08-31.