inspect-robots CLAUDE.md

inspect-robots CLAUDE.md is an instructions file for coding agents from robocurve/inspect-robots. It costs 2,035 tokens per session, scanned A, original, MIT.

Repository instructions for Inspect Robots, an open-source framework for evaluating vision-language-action models on robots or simulated robots. These models use observations to choose actions, while a robot or simulation executes them.

In plain words
What is it for?
Use it when navigating or changing the Inspect Robots framework, its evaluation flow, mock robot world, tests, or design plans.
Why use it?
It explains the project structure, core concepts, testing approach, and which related work belongs in separate repositories.

Instructions file

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 instructions/robocurve/inspect-robots/claude-md
Clone the repo
git clone --depth 1 https://github.com/robocurve/inspect-robots

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 inspect-robots CLAUDE.md

README.md
[![agentmods](https://agentmods.dev/badge/instructions/robocurve/inspect-robots/claude-md.svg)](https://agentmods.dev/instructions/robocurve/inspect-robots/claude-md)
Your own site
<a href="https://agentmods.dev/instructions/robocurve/inspect-robots/claude-md"><img src="https://agentmods.dev/badge/instructions/robocurve/inspect-robots/claude-md.svg" alt="Measured on agentmods" height="20"></a>
Per session 2,035 This file is loaded in full into every session.
When invoked 2,035 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 $0.02035 $0.02035
Opus 5 $0.01018 $0.01018
Sonnet 5 $0.00407 $0.00407
Haiku 4.5 $0.00203 $0.00203

Measured 5d ago against content hash 6588af02f4dc, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

inspect-robots CLAUDE.md 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 5d 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.

CLAUDE.md · 128 lines

How it starts

The opening of the file, as written. The whole thing — 128 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Inspect Robots — agent guide

Inspect Robots is the "Inspect AI for robotics": an open-source evaluation framework for physical AI / VLA (vision-language-action) models. This repo is the framework; concrete benchmarks and backend adapters live elsewhere (see below).

The one big idea

LLM evals have one swappable input (the model). Robotics evals have two:

  • Policy (the VLA "brain") — observation → ActionChunk (open-loop horizon).
  • Embodiment (the robot/sim "body + world") — executes actions, owns spaces.

A Task (a dataset of Scenes + scorers) is defined independently of both. eval() checks a (policy, embodiment) pair is compatible, runs the closed-loop rollout, scores it, and writes an immutable EvalLog. Mirrors Inspect AI's Task = dataset + solver + scorer, eval(), EvalLog, registry/decorator model.

Layout

  • src/inspect_robots/ — the package (see src/inspect_robots/CLAUDE.md for the module map).
  • tests/ — pytest; the CubePick mock world exercises the whole stack with no hardware or sim.
  • plans/ — design docs. plans/0001-foundation-design.md is the authoritative spec (read its §9–§11 "binding resolutions" before changing core interfaces).
  • examples/ — runnable demos (quickstart.py).
  • plugins/* — first-party plugin packages (concrete sims/VLAs that are out of scope for the numpy-only core), each its own package with its own pyproject, entry point, tests, and coverage scope. A uv workspace ([tool.uv.workspace]) ties them in: uv sync --all-packages --extra dev installs core + all plugins editable. They never count toward the core 100% gate (coverage is scoped to inspect_robots). E.g. plugins/inspect-robots-isaacsim/ (Isaac Lab embodiment), plugins/inspect-robots-ros/ (ROS 1/ROS 2 embodiment speaking rosbridge with no ROS dependency), plugins/inspect-robots-xpolicylab/ (policy adapter speaking the XPolicyLab websocket protocol — 40+ served VLAs, no xpolicylab dep), and plugins/inspect-robots-agent/ (LLMs as policies via the OpenAI-compatible wire formats or Anthropic's native Messages API — httpx only, no provider SDKs; registered as agent), and plugins/inspect-robots-capx/ (CaP-X code-as-policy over perception and IK HTTP servers; registered as capx), and plugins/inspect-robots-voice/ (local microphone transcription with Parakeet and local Kokoro policy narration; registered as the voice operator input and speaker sink).

Read the full file on GitHub · 128 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. 5d ago First seen · 128 lines · 2,035 tokens per session scan A 6588af02f4dc

Subscribe to this mod's changes

inspect-robots CLAUDE.md is an instructions file published in the GitHub repository robocurve/inspect-robots (205 stars, last pushed 2d ago), licensed MIT. It adds 2,035 tokens to every session, about $0.0102 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.