Skip to main content

AutoBuild

Coding assistants can work through a backlog, but the result becomes hard to trust when each assistant carries a different build process in its prompt. AutoBuild puts the campaign in one Python application so the same sequence, checks and delivery rules apply whichever supported assistant runs it.

A fresh builder changes one ready item from the project's tracked queue in an isolated Git worktree. The queue can be Pinax or a supported BACKLOG.md table. A fresh reviewer sees the approved request, changed files and test evidence without the builder transcript. The application accepts, parks or stops the item, then records enough evidence to judge the outcome without reconstructing a chat session.

The workflow is harness-neutral. Claude Code, Codex and GitHub Copilot sit behind adapters selected at startup, and an optional Claude Agent SDK lane can take seats on an API key when a subscription lane reaches its usage stop level; the campaign sequence does not change with the assistant, operating system or shell. The project profile names the model for each seat and, optionally, the effort level that controls how much that model thinks; each adapter passes both in its assistant's own form.

Run sequence

  1. Check the selected harness, repository and project test command before claiming work.
  2. Build and review each ready item in its own isolated worktree. Independent items build at the same time, as many as the host allows; set concurrency = 1 in the profile to build one at a time.
  3. Deliver accepted product and tracker commits to the selected branch mode. Local PR delivery keeps the result on the invoking branch; protected delivery merges and pushes the default branch. Accepted items deliver one at a time.
  4. Continue until the queue is dry, the item limit is reached, or one hold on new work lasts the hold limit. An error in one item ends only that item.

Repository contents

  • src/autobuild/: the cycle and its adapter interfaces
  • src/autobuild/adapters/: the adapters, one per mechanism
  • src/autobuild/enforcement/: gates, validators and schemas
  • skills/autobuild-plan/: plans, reviews and registers an end-to-end queue, then stops before launch
  • skills/autobuild/: configures and launches the Python workflow against a ready queue
  • skills/autobuild-coordinated/: runs the same method inside a coding-assistant session with fresh builder and blind reviewer seats, chosen at launch instead of the application
  • tests/: the test lane

Start here

The setup and run guide explains installation, harness authentication, item briefs, Pinax and BACKLOG.md, the project profile, campaign results, refill and failure handling from a standing start.

AutoBuild has two stages. Use autobuild-plan for research, planning, independent review and tracker registration. After the owner approves that result, use autobuild to run the queue. The operating guide starts with the planning stage.

Release 0.9.1 is available from PyPI as autobuild-factory and as a Python wheel and source archive on the GitHub release page. The guide also has a macOS setup path and GitHub Copilot setup.

Platform and coding assistant are separate choices. You can use Codex on macOS, GitHub Copilot on Windows, or any other supported combination.

The architecture guide explains the layers, state machines, ports, adapters, evidence chain, Git delivery model and extension points.

Run the test lane with uv run --native-tls python -m pytest tests -q. Add -n 8 --dist loadfile to run it on eight pytest-xdist workers.

The harness adapter guide explains how one builder and reviewer contract runs through Claude Code, Codex and GitHub Copilot.

Delivery checks

AutoBuild probes its adapters before claiming work, confines changes to an isolated worktree, and runs only the declared validator. It verifies the remote revision for protected delivery or an explicitly pushed current branch. An accepted result must carry matching diff, validator, review, product-commit and tracker-commit evidence.

Scope and limits

Product decisions, public releases and production deployments stay outside AutoBuild. An optional queue-refill plan records proposed work and unresolved questions. It cannot make a proposal runnable.

Metadata

Release files for autobuild-factory 0.9.1

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

Source distribution (sdist)

Source distribution for autobuild-factory 0.9.1
File Size Uploaded Provenance Actions
autobuild_factory-0.9.1.tar.gz 438.8 kB With provenance Details for autobuild_factory-0.9.1.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for autobuild-factory 0.9.1
File Interpreter ABI Platform Actions
autobuild_factory-0.9.1-py3-none-any.whl Python 3 none any Details for autobuild_factory-0.9.1-py3-none-any.whl

Total release size: 652.4 kB

Release files / autobuild_factory-0.9.1.tar.gz

Download URL autobuild_factory-0.9.1.tar.gz
Size 438.8 kB
Tags Source
SHA-256 checksum
How to use checksums
f2a1eda1c148b10744be3d23c4fd0462375c859d97d3a9afebb8ca6c28a52bee
BLAKE2b-256 checksum
How to use checksums
d0e3d0559f491144ae47301026d2cfae59522ea3ee4a20704d154d95e4b169c8
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

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.

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 Oct 26, 2026.

Transparency log

Release files / autobuild_factory-0.9.1-py3-none-any.whl

Download URL autobuild_factory-0.9.1-py3-none-any.whl
Size 213.6 kB
Tags Python 3
SHA-256 checksum
How to use checksums
ba7c8d64b42f872f041326d78716d2125fab864fe1f2258eb5ad8c0f92147439
BLAKE2b-256 checksum
How to use checksums
7aedbee7a81c2de5e2f1cc9944b047d4cdf82f99e91732b0e34f76185024466a
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

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.

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 Oct 26, 2026.

Transparency log

Security for autobuild-factory 0.9.1

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 autobuild_factory-0.9.1.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 Oct 26, 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.

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