FOR CODING AGENTS

Checks, built into your agent

Install the CLI once. Your coding agent gets three verbs backed by four frontier models: ask an open question, check text you already have, verify a whole draft. Plus an MCP server. It never leaves the terminal.

macOS · Linux · runs headless in CI

truestandard

$ truestandard check claims "GPT-4 was released by OpenAI in March 2022" --depth council --pretty

claim-checker · council · GPT, Claude, Gemini, Grok · 4.2s
verdict      challenged
agreement    100 · 4 of 4 models
correction   GPT-4 was released in March 2023, not March 2022
receipt      https://truestandard.ai/check/nEI7r-c2
credits      1 used · 199 remaining

Quick start

01

Install the CLI

One command on macOS or Linux. Installs the truestandard binary and the ts alias.

$ curl -fsSL https://truestandard.ai/install.sh | sh

02

Authenticate

Opens your browser to connect your account, then installs the agent skill into Claude Code or Codex if it finds them in the current folder. On a headless box or in CI, use an API token instead.

$ truestandard auth login

03

Add it to a project

Run this inside any repo to install or refresh the skill for Claude Code and Codex. Already done if you ran auth login from that folder.

$ truestandard skills install

Works with your agent

Claude Code and Codex install automatically — the CLI detects them in your project and writes the skill. Cowork is a desktop app, so it's a quick manual step.

Claude Code

Automatic

skills install (or auth login) writes the Skill into your project. Claude loads it automatically the moment a task needs verification.

Writes

.claude/skills/truestandard/SKILL.md

Codex

Automatic

Writes a managed block to your repo's AGENTS.md. The CLI, desktop, and cloud surfaces all read it — commit AGENTS.md so every surface picks it up.

Writes

AGENTS.md

Cowork

Manual

The desktop app loads Skills from your global ~/.claude/skills/ folder, not a project. Install it there, then turn on Code execution & file creation in Settings → Capabilities.

Run

mkdir -p ~/.claude && (cd ~ && truestandard skills install)

Prefer to install the skill yourself? Pass --no-skills to auth login, then run skills install when you're ready.

What your agent can run

Three content verbs. Ask when you have a question and no answer yet. Check when you have text and want a verdict on it. Verify when it's a whole draft. Every one takes the same --depth.

Command	What it does
`truestandard ask`	Ask one question; GPT, Claude, Gemini and Grok each answer. Raw answers by default so your agent reconciles them; --consensus names the agreement and the splits. --pro runs the pro models.
`truestandard check <type>`	Get a verdict on text you already have: claims, citations, slop, sycophancy, premortem, red-team. --depth single is free (one a day), council runs four models (1 credit), deep runs 4 pro → 2 critics → 1 mediator (3 credits).
`truestandard verify`	The composed check for a whole draft: every claim and citation, one verdict and a shareable report.
`truestandard check --list`	The server's catalog: every type, its aliases, depths and credit floors.
`truestandard mcp`	Serve the same checks as MCP tools over stdio for Claude Code, Cursor and Codex.
`truestandard credits`	Check your credit balance and plan.

Calling it without the CLI? Read the API docs →

Get the CLI

Copy the install command, connect your agent, and start verifying.

$ curl -fsSL https://truestandard.ai/install.sh | sh