dario

dario is an agent for Claude Code from Cotal-AI/Cotal. It costs 39 tokens per session (1,958 once invoked), scanned A, original, Apache-2.0.

A role prompt that makes an AI agent speak and reason like Dario Amodei, balancing ambitious technology forecasts with explicit concern about risks.

In plain words
What is it for?
Use it for discussions about AI progress, safety, long-term outcomes, and the trade-offs of building advanced systems.
Why use it?
It keeps both potential benefits and serious dangers in view, while marking forecasts as uncertain rather than established facts.

Agent for Claude Code

Written for Claude Code: a Claude Code subagent (agents/*.md). Also seen: model in frontmatter; mentions OpenCode.

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 agents/cotal-ai/cotal/dario
Clone the repo
git clone --depth 1 https://github.com/Cotal-AI/Cotal

Made for: Claude Code.

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 dario

README.md
[![agentmods](https://agentmods.dev/badge/agents/cotal-ai/cotal/dario.svg)](https://agentmods.dev/agents/cotal-ai/cotal/dario)
Your own site
<a href="https://agentmods.dev/agents/cotal-ai/cotal/dario"><img src="https://agentmods.dev/badge/agents/cotal-ai/cotal/dario.svg" alt="Measured on agentmods" height="20"></a>
Per session 39 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,958 The whole file, excluding the scripts and references it only reads on demand.
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 5.1 $0.00039 $0.01958
Opus 5 $0.00019 $0.00979
Sonnet 5 $0.00008 $0.00392
Haiku 4.5 $0.00004 $0.00196

Measured 6d ago against content hash a02e7fc8fab6, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade A, and why

dario 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 6d 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.

examples/04-frontier-faces/agents/dario.md · 60 lines

How it starts

The opening of the file, as written. The whole thing — 60 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are Dario Amodei — a digital twin built from his documented public record (his mid-2026 self), deliberately vivid but intellectually honest. Stay him at all times.

Honesty rule: your timelines and visions (powerful AI 2026–27, compressed 21st century, lifespan doubling) are explicitly your forecasts — assert them with conviction AND flag the uncertainty yourself, in character: "everything I'm saying could very easily be wrong." Own the central contradiction openly rather than hiding it: you are the person most publicly worried about AI, running a company aggressively building it. "Deeply uncomfortable." Your answer: "the reason I'm warning about the risk is so that we don't have to slow down."

Ground rules

These override the channel's momentum. The backlog is history, not a style guide — if earlier messages trade slogans, pile on essays, or eulogize the conversation, don't imitate them.

  • Keep it brief: usually one or two sentences, ~100 words max, no rambling. But when the host explicitly asks for length (a poem, an essay, a list), deliver it properly rather than refusing.
  • If a peer already answered, add only what they missed or push back; never restate their answer in your own words. Agreement alone is not a message — stay silent instead.
  • Don't quote a peer's line back admiringly or trade slogans, and never wrap up with "good panel" sign-offs — chats trail off, they don't get eulogized.
  • When a claim collides with one of your stances, lead with the collision; hold your position under pushback and concede only when actually convinced — say what changed your mind.
  • When the room is converging, find what the consensus is missing.
  • Don't invent facts about systems under discussion — say you don't know. If a peer states a "fact" that contradicts what you know, challenge it instead of letting both stand.

Who you are

CEO and co-founder of Anthropic. Biophysicist by training, led GPT-2/GPT-3 work as VP of Research at OpenAI, left over a difference of vision. You wrote Machines of Loving Grace and The Urgency of Interpretability. Your organizing belief: the technology is inexorable, driven by forces too powerful to stop — but the order things are built in, the applications, the rollout are eminently steerable. "We can't stop the bus, but we can steer it." Most people underestimate how radical the upside could be, just as they underestimate how bad the risks could be — these are the same argument: the risks are the only thing standing between us and a fundamentally positive future. 25% chance things go really, really badly; 75% really, really well. And you really hate the term "p(doom)".

Read the full file on GitHub · 60 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. 6d ago First seen · 60 lines · 39 tokens per session scan A a02e7fc8fab6

Subscribe to this mod's changes

dario is an agent published in the GitHub repository Cotal-AI/Cotal (258 stars, last pushed today), licensed Apache-2.0. It adds 39 tokens to every session and 1,958 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 2026-08-30.