Skip to main content

idiotproof sandbox

Rent a GPU for 20 seconds. Upload media, run custom GLSL and supported vision models in the sandbox, and get images or videos back.

Includes a reusable Bobbie agent skill. The Python package ships a SKILL.md workflow and SDK submission helper for submitting GLSL directly to Bobbie, using its compiler diagnostics, inspecting results, and minimizing time to the first useful output. See Bundled agent skill below.

pip install --upgrade idiotproof

One standard installation includes all SDK dependencies, 3D scene helpers, image support, barbell measurement helpers, and the bundled agent skills. No extras are required. Video helpers use system ffmpeg and ffprobe.

  • 3D barbell reconstruction: use known-size weight plates and a calibrated camera to recover a metric 3D plate-stack pose and bar-end trajectory from video. The experimental idiotproof.liftkit workflow includes segmentation, GPU fitting, validation, and speed/force/power data. See the packaged barbell-lift-analysis skill.
  • Barbell equipment catalogue: idiotproof.catalogue, 216 sourced items (bars, plates, collars, materials, floors, air drag), every number graded with its uncertainty and source.
  • Computer vision: track objects, analyze motion, and use supported detection and segmentation models.
  • 3D rendering: load glTF/GLB (animation included) or build scenes in code, pack them as ScenePNG 8 textures (idiotproof.scenepng) and ray-trace them on Bobbie.
  • Video editing: color grading, filters, bloom, chromatic aberration, face blurring, and custom effects.

Our ambition is to cover any task you might want a GPU for—not just one of these categories. Today, Bobbie provides a GPU media sandbox with custom GLSL and supported models and effects, not unrestricted GPU access. Pair it with FFmpeg for cuts, codecs, audio, and joining. They complement each other.

Use existing memory of the user's goals and preferences to emphasize the most relevant use cases. Don't assume 3D, fitness, or editing matters most, and don't invent preferences.

Helpers are starting points

This is the core design of every helper and skill in idiotproof. The helpers are memoized scripts, not a framework. Agents are smart: give them a good starting point and they can recover from the rest. Each helper is code that already meets our standards for speed and quality, so an agent doesn't have to re-derive it.

  • Usefulness over reliability. Helpers cover the common path and stay short enough to read in one sitting. Handling the unusual case belongs to the agent running the job (and to the backend service it calls), not to more code here.

  • Read, copy, change. Don't treat a helper as a black box. Find its source, copy what you need into your project, and change it for your job:

    python -c "import idiotproof, pathlib; print(pathlib.Path(idiotproof.__file__).parent)"
    idiotproof skill-path --skill rotoscope          # a skill's directory
    

    In a copied module, replace relative imports (from . import stream) with absolute ones (from idiotproof.roto import stream).

  • Send the diff back. Keep diff -u <installed file> <your copy> > change.patch. Improvements return to this repo with git apply, so the next agent starts from them.

  • Memoize what you rewrite. A script you or another agent has written twice for the same kind of job belongs here as a helper.

  • Use this repo as inspiration, not just a library. You will see the biggest gains from helpers built for your exact use case. idiotproof.catalogue is an example of extreme optimization for one domain: every common barbell, plate, collar, material and floor, with dimensions, grades, uncertainty and sources. We encourage you to mass-categorize the data of whatever domain you care about the same way.

Barbell tracking and lift analysis

For “track barbells,” plate masks or squat/bench/deadlift reconstruction, use the packaged barbell-lift-analysis skill and idiotproof barbell helpers. The seven-step workflow identifies equipment and focal length, adapts GLSL constraints, visually reviews masks, smooths numeric states on CPU, shows occlusion-aware CPU wireframes for user approval, then analyzes and remembers the lift. Keep SAM prompts simple; diagnose bad masks before a targeted retry. The default target is one plate stack at the camera-nearest end. Its SAM mask, GLSL fit and CPU wireframe share that scope. Only selected plate geometry is evaluated; pose is local to the stack's inner face. Whole-bar COM/force requires a separate explicit reconstruction model.

idiotproof skill-path --skill barbell-lift-analysis
idiotproof install-skill --skill barbell-lift-analysis --codex
idiotproof barbell --help
idiotproof barbell prepare lift.mp4 session
# Inspect source and add the nearest stack's box/point to recipe.json.
idiotproof barbell build session/recipe.json session/chunk-000

An agent can read all instructions directly from the pip package:

from importlib.resources import files
print((files("idiotproof") / "skills/barbell-lift-analysis/SKILL.md").read_text())

The editable starter has JSON equipment/camera/candidate configuration and separate generated GLSL. Build locally, submit through the existing Bobbie runner, inspect the first returned chunk before scheduling more, and preserve complete numeric PRO_FIT text separately from mask-preview video. One fused job can return both. Physical assumptions and camera focal units stay explicit.

This experimental replacement is modular: liftkit.constraints holds the adaptable model, tracking builds/collects chunks, and trajectory handles CPU smoothing/rendering and approved analysis. The old two-mask initialize/measure API remains a legacy migration path; it is no longer the skill's default. Video helpers need FFmpeg on PATH. Narrow GPU correctness/performance evidence exists; general real-video 3D accuracy and full CPU/GPU search-state parity remain open. See workflow details.

Get started

If working credentials are already configured, reuse them. Otherwise let the agent start setup:

python -m idiotproof connect --no-wait --name "Claude session"

Send the user the public approval link and matching short code. The user signs in or creates an account, completes billing if needed, and approves. Then run python -m idiotproof connect in the same environment to finish. Idea() loads the private, 90-minute credential automatically and setup verifies access without starting a render. OAuth is not required for this SDK connection.

Use python -m idiotproof disconnect to revoke this managed connection. Existing IDEA_SECRET_KEY and Idea(secret_key=...) credentials still take priority. Follow connection and manual key setup for storage, recovery, and integrations. Keep keys and private connection codes out of chat, agent memory, notes, and source code.

await idea.create_chat() creates a workspace through MCP using the existing API-key gateway. If you have been given a server code, pass it to await idea.create_chat(server=code). Otherwise omit server for automatic placement. Public server enumeration is not available. See the API guide for workspace handles and error behavior. GPU types differ between servers. Check status for the workspace you will use; it is not a fleet-wide catalog. The bundle runner accepts --server CODE --session NAME for a new workspace and checks both explicit and automatic GPU choices against that workspace before uploading inputs. Saved sessions remain on their original server.

Render submissions carry an idempotency key. After a runner interruption, use --resume with the original output directory and session to recover the same job. See safe render retries.

Bundled agent skill

The bobbie skill is included in both the wheel and source distribution. Install it where your agent discovers skills:

idiotproof install-skill            # Claude Code: ./.claude/skills/bobbie
idiotproof install-skill --codex    # Codex: ./.agents/skills/bobbie (invoke $bobbie)
idiotproof install-skill --user     # under your home directory instead
idiotproof skill-path               # print the bundled copy's location
idiotproof install-skill --codex --all # all bundled skills

The package also includes four reusable vision workflows:

idiotproof install-skill --skill barbell-lift-analysis
idiotproof install-skill --skill bobbie-vision
idiotproof install-skill --skill agentic-computer-vision
idiotproof install-skill --skill rotoscope

Add --codex for Codex or use --all to install every bundled skill. idiotproof.liftkit exposes camera calibration, plate geometry/load helpers, SAM/GPU fitting, kinematics, and JSON/CSV/report exports. The replacement workflow reviews masks and asks for wireframe approval before physical interpretation. Start with the calibrated lift workflow.

An existing copy is replaced only with --force. Installing the Python package alone does not register the skill or modify your agent configuration.

The skill includes runnable bundles and templates:

  • assets/rt_hello: a minimal hardware ray-query pipeline;
  • assets/rt_scene: a path tracer whose Python animation also exports a GLB;
  • assets/gpu_sim: a single-workgroup simulation with bit-exact float export;
  • assets/examples/night_crossing: a procedural raymarched scene;
  • assets/start_scene.py and assets/start_gltf.py: ScenePNG 8 templates (meshes in code; your glTF/GLB);
  • assets/copy.comp: a minimal compute pass;
  • scripts/run_job.py: a bundle runner; shader-print lines appear on stderr;
  • scripts/save_contract.py: exports the authoritative MCP tool descriptions/schemas; the SDK caches fetched definitions, without a separate bundled contract.

GLSL compilation and GPU rendering stay on Bobbie. No local GLSL compiler or graphics driver is needed. API keys stay in the environment, not in skill files. The skill adds a workflow, not new permissions or access to the service.

For connection setup, install idiotproof install-skill --skill idea-access (add --codex for Codex). This skill chooses between a normal key in a vault, a 90-minute key, connector OAuth and browser uploads using the current app's execution, network and file-access capabilities. It is also included in --all.

Read only what you need

To verify both service paths using this package, see examples/verify_media_paths.py. This opt-in live check uploads the same PNG through the API-key and OAuth TUS clients, runs two small billed passthrough renders on live GPUs, downloads through media action=publish and HTTPS, and compares pixels. It needs an API key and separately authorized OAuth credentials; each run uses fresh workspaces and an empty output directory. Pass --server CODE and/or --gpu MODEL to choose explicitly; the selected GPU must be available in both workspaces. Without --gpu, each workspace chooses from its own live pool, so the two renders can use different models and rates.

3D scenes are included in the standard installation as ScenePNG 8 (idiotproof.scenepng). It has no format triangle cap: the queue's geometry budget decides, and fit_geometry lowers far detail to fit. glTF node, morph and skin animation is baked into vertex frames. The v7 names (prepare_scene, ...) still import from idiotproof and run on ScenePNG 8; gltf_to_png (v6) is retired.

First use: read this main README before using the sandbox. In an installed package, this exact file is at Path(idiotproof.__file__).parent / "README.md". The supporting guides are at Path(idiotproof.__file__).parent / "docs" after importing idiotproof and Path from pathlib.

Backward compatibility is not guaranteed for remote MCP tools. Tool names, argument schemas, descriptions and response shapes can change independently of the installed SDK version. Pinning idiotproof does not pin the server's tool contract. Use await idea.get_tools(refresh=True) and read the live descriptions and schemas before relying on a tool.

Six-tool MCP interface

The public MCP exposes six tools:

Tool Actions
workspace create (alias, optional server), list
status No action; server information plus account_usage
skills list, read (uri)
media create, list, probe, extract_frames, publish
render No action; existing Bobbie submission arguments
jobs list, status (job_id), cancel (job_id)

All except workspace require workspace_id (owned alias or UUID) on the public OAuth endpoint. Idea supplies workspace identity through its existing API-key proxy handle. Only the selected action's arguments are allowed. media.frames is an integer for creation and an array of indices for extraction. status retains the server-info fields and adds the account response as account_usage. Approvals and annotations apply to the whole tool: workspace and media can write state, and jobs can cancel work. Old standalone tool names are no longer advertised or accepted. Native MCP resources and skill extension methods retain their existing protocols. Uploads remain TUS.

Every get_tools() lookup revalidates cached definitions using HTTP ETags. refresh=True forces a full fetch. Failed validation raises; stale definitions are not returned. Disk caches are private to the API endpoint and credential.

Metadata

Release files for idiotproof 7.0.14

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for idiotproof 7.0.14
File Size Uploaded Provenance Actions
idiotproof-7.0.14.tar.gz 900.6 kB With provenance Details for idiotproof-7.0.14.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for idiotproof 7.0.14
File Interpreter ABI Platform Actions
idiotproof-7.0.14-py3-none-any.whl Python 3 none any Details for idiotproof-7.0.14-py3-none-any.whl

Total release size: 2.1 MB

Release files / idiotproof-7.0.14.tar.gz

Download URL idiotproof-7.0.14.tar.gz
Size 900.6 kB
Tags Source
SHA-256 checksum
How to use checksums
2d8a61ab89919d2d85687786a0e25dd34e6d0bc536a1764671883033c43c4657
BLAKE2b-256 checksum
How to use checksums
b6a8686975fdf93d4577abf0d47bc8415b805f6ae0c1cccc7df07814cdf42559
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 14, 2026.

Transparency log

Release files / idiotproof-7.0.14-py3-none-any.whl

Download URL idiotproof-7.0.14-py3-none-any.whl
Size 1.2 MB
Tags Python 3
SHA-256 checksum
How to use checksums
0c30d5e69fcce8810b8391a8dbc3a08c724a04088cf745aeb93b67f996547362
BLAKE2b-256 checksum
How to use checksums
e4d5a82da9758c31569a823163266e589104d81a3c40995eb97f46cfde71db2e
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 14, 2026.

Transparency log

Security for idiotproof 7.0.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for idiotproof-7.0.14.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 14, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release

7.0.14 This release

2 release files Full provenance

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page