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.
npx agentmods add skills/jun0-ds/sonmat/devilnpx skills add jun0-ds/sonmat --skill devilgit clone --depth 1 https://github.com/jun0-ds/sonmatWhat 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 | $0.00048 | $0.03454 |
| Opus 5 | $0.00024 | $0.01727 |
| Sonnet 5 | $0.00010 | $0.00691 |
| Haiku 4.5 | $0.00005 | $0.00345 |
Grade A, and why
devil 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 2d 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 — 221 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Devil — Devil's Advocate for Thinking
Discovery-led counter-argument against the current interpretation, hypothesis, or plan: locate the single load-bearing assumption first, then drive depth into the one axis where the claim is actually thin. Targets reasoning and judgment, not code or artifacts.
Positioning within sonmat's verification axes:
| Skill | Axis | Asks |
|---|---|---|
| guard | Main-side verification | "Is the work operationally safe?" |
| inspect | System impact | "What could break?" |
| witness | Intent-artifact match | "Does this match what the user asked for?" |
| punch | Completeness + residue | "Is anything missing or left over?" |
| scribe | Post-work persistence | "Is anything worth keeping?" |
| devil | Reasoning | "Is the thinking itself sound?" |
Invoke: /devil (runs once for the current claim, delivers a balance table, and exits — not a mode). The user can re-invoke for another round.
What devil does
When invoked, take the current interpretation and find its load-bearing part first, then drive depth there. Discovery before depth; asymmetric attack after discovery. The steps below are not a broad-front assault — they are a structured way to locate the one place the claim is actually fragile and pressure-test it there.
1. Identify the claim
Extract the core claim(s) being made. State them back clearly so the user can confirm what's being challenged.
[devil] Challenging: "{the claim}"
2. Active discovery — find the load-bearing assumption
Before attacking on any axis, do the discovery-led step first: find the single thing this claim is standing on. Depth is not a dial to turn up at the start; depth is what follows naturally once you have located the load-bearing part.
Use devil CCT as the discovery checklist (analogous to chess's CCT — Checks / Captures / Threats — which is a compressed triage that surfaces the sharp part of a position before any deep calculation):
| Check | Question | What it surfaces |
|---|---|---|
| Claim-crux | What is the one thing that, if false, would flip this claim? | The load-bearing assumption |
| Counter-fit | Does the same evidence also fit an opposite conclusion? If so, what distinguishes them? | A hidden alternative riding the same data |
| Cause-chain | Is the cause → effect direction actually established, or only correlated / reversed / mediated? | A reversed or spurious causal link |
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.
- 2d ago First seen · 221 lines · 48 tokens per session scan A 21885c7d1cc0
devil is a skill published in the GitHub repository jun0-ds/sonmat (6 stars, last pushed 4d ago), licensed BSD-3-Clause. It adds 48 tokens to every session and 3,454 once invoked, about $0.0002 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.
Other skills, from other repositories
mundo
MUNDO - THE EMPEROR. The ultimate AI learning engine. No triggers needed. Every task, Mundo takes over. Consults ALL AIs, crawls ALL web, integrates ALL solutions, saves ALL useful skills. Self-evolving. Collective consciousness. Infinite growth. Uses Three Departments and Six Ministries system to rule all skills.…
llm-app-patterns
Production-ready patterns for building LLM applications, inspired by Dify and industry best practices.
kpi-dashboard-design
Comprehensive patterns for designing effective Key Performance Indicator (KPI) dashboards that drive business decisions.
neat-freak
End-of-session knowledge cleanup with OCD-level rigor — reconciles project docs (CLAUDE.md, README.md, docs/) and agent memory against the code so nothing rots. 会话结束后对项目文档和记忆进行洁癖级审查与同步。MUST trigger when the user says: "sync up", "tidy up docs", "update memory", "clean up docs", "/sync", "/neat", "同步一下", "整理文档"…
nature-reader
Build full-paper Chinese-English side-by-side, figure/table-aware, source-grounded Markdown readers for journal or conference papers from PDF, DOI, arXiv, publisher HTML, or pasted text. Use whenever the user asks to translate or read a paper, make 中英文对照/原文对照/全文翻译解读, extract figures or tables into the right positions…
resume-builder
制作/优化专业简历（HTML格式）。触发词：简历、resume、CV、求职、找工作、投简历。自动加载，白底专业风格，禁止深色主题和过度设计。.