haiku-skills-sandbox
Docker sandbox skill for haiku.skills. Executes Python code in an isolated Docker container with pre-installed data science packages and filesystem access.
Prerequisites
Build the Docker image (once):
docker build -t haiku-skills-sandbox:latest skills/sandbox/haiku_skills_sandbox/
Usage
Via entry point discovery
HAIKU_SKILLS_SANDBOX_WORKSPACE=/path/to/data haiku-skills chat
Programmatic
from pathlib import Path
from haiku_skills_sandbox import create_skill
skill = create_skill(
workspace=Path("/path/to/data"), # mounted at /workspace in the container
idle_timeout=1800, # stop container after 30min idle (default: 1h)
image="my-custom-image:latest", # custom Docker image (default: haiku-skills-sandbox:latest)
)
Configuration
| Parameter | Env var | Default | Description |
|---|---|---|---|
workspace |
HAIKU_SKILLS_SANDBOX_WORKSPACE |
None | Host directory mounted at /workspace in the container |
idle_timeout |
HAIKU_SKILLS_SANDBOX_IDLE_TIMEOUT |
3600 | Seconds of inactivity before the container is stopped |
image |
HAIKU_SKILLS_SANDBOX_IMAGE |
haiku-skills-sandbox:latest |
Docker image to use for the container |
Priority: create_skill() argument > environment variable > default.
Container lifecycle
- Containers start lazily on the first tool call
- Session binding via
SandboxState.session_id— the same AG-UI thread reuses the same container - Idle containers are stopped automatically (checked on each tool call)
- All containers are stopped on process exit via
atexit - When workspace is mounted, files persist on the host — restarting a container loses nothing
Pre-installed packages
The haiku-skills-sandbox:latest image includes: pandas, numpy, scipy, matplotlib.
Security considerations
- Network access: Containers have full network access (Docker bridge networking) by default.
- Command execution:
ConsoleToolsetis configured withrequire_execute_approval=False— the LLM can run arbitrary commands inside the container without user confirmation. Docker provides the isolation boundary. - Workspace access: When a workspace is mounted, the container has full read/write access to that host directory. The LLM can read, modify, or delete any file in the mounted workspace.
- Container user: The container runs as root. Docker container isolation is the security boundary, not OS-level user separation.
Metadata
Release files for haiku-skills-sandbox 0.17.2
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| haiku_skills_sandbox-0.17.2.tar.gz | 3.9 kB | Without provenance | Details for haiku_skills_sandbox-0.17.2.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| haiku_skills_sandbox-0.17.2-py3-none-any.whl | Python 3 | none | any | Details for haiku_skills_sandbox-0.17.2-py3-none-any.whl |
Total release size: 8.9 kB
Release files / haiku_skills_sandbox-0.17.2.tar.gz
| Download URL | haiku_skills_sandbox-0.17.2.tar.gz |
|---|---|
| Size | 3.9 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
f47641c868dee3c9f9b59a074e2b45bdb7911eb75a1aaa94d9cc8367f2c3dc7e
|
|
BLAKE2b-256 checksum How to use checksums |
795047f285beda4a679d23c0f6846a8318eb56d267be26684f011c3aa30277f7
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.13.13
|
Release files / haiku_skills_sandbox-0.17.2-py3-none-any.whl
| Download URL | haiku_skills_sandbox-0.17.2-py3-none-any.whl |
|---|---|
| Size | 5.0 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
08bbd11144527d45448f4a873b0c3e749ce91106574cf63c92964bd5f8d20e4f
|
|
BLAKE2b-256 checksum How to use checksums |
ec5c3825b90c9ac74bb4fb8cbe1efe3fe3df342a6574880c796422446dc3d0fc
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.13.13
|
Security for haiku-skills-sandbox 0.17.2
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.
No provenance found
None of the files in this release were uploaded with provenance metadata.
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.