Skip to content

Add MaxText automated release preparation skill#3910

Draft
igorts-git wants to merge 1 commit into
mainfrom
add-release-skill
Draft

Add MaxText automated release preparation skill#3910
igorts-git wants to merge 1 commit into
mainfrom
add-release-skill

Conversation

@igorts-git
Copy link
Copy Markdown
Collaborator

@igorts-git igorts-git commented May 14, 2026

Description

This PR introduces a standardized AI agent skill file (SKILL.md) designed to orchestrate the end-to-end validation and release preparation lifecycle for MaxText PyPI packages.
The workflow automates tedious verification matrices while enforcing strict zero-regression safety boundaries before prompting a maintainer for final publication authorization.

Key Workflow Constraints & Features

  • Strictly Read-Only Validation: Prohibits automated source code patches during execution. If any test returns unexpected stderr, a stack trace, or a non-zero exit, the agent halts and reports verbatim diagnostics.
  • Clean State Enforcement: Enforces fresh virtual environments (uv venv) for isolated phases and validates clean git tree boundaries to prevent stale artifact pollution.
  • Phased Execution Matrices:
    • Phase 1: Version verification (src/maxtext/__init__.py) and PyPI wheel compilation.
    • Phase 2: Single-host TPU validation (Pre-training, Post-training SFT/GRPO, and vLLM inference decodes).
    • Phase 3: Multi-host cluster orchestration (Image builds, McJAX/Pathways XPK job scheduling, checkpoint persistence checks).
    • Phase 4: Source checkout integrity builds and benchmark runner validation matrices (Stable vs. Nightly).
  • Maintainer Ownership: Generates a final executive report matrix detailing successful logs vs. anomaly snippets to support the final human GO/NO-GO release decision.

Tests

Validation steps:

  • I created a new TPU v5 VM.
    • Installed required OS patches
    • Configured docker
    • Installed gemini-cli
    • Cloned this maxtext branch.
  • Opened "gemini" in the maxtext directory, signed in, and prompted it: "please run the release process"
  • Gemini-CLI asked me a few questions (HF_TOKEN, XPK cluster name, etc)
  • I clicked Ctrl+Y to enable the YOLO mode. It ran for a few hours processing all validation steps.

Checklist

Before submitting this PR, please make sure (put X in square brackets):

  • I have performed a self-review of my code. For an optional AI review, add the gemini-review label.
  • I have necessary comments in my code, particularly in hard-to-understand areas.
  • I have run end-to-end tests tests and provided workload links above if applicable.
  • I have made or will make corresponding changes to the doc if needed, including adding new documentation pages to the relevant Table of Contents (toctree directive) as explained in our documentation.

@igorts-git igorts-git force-pushed the add-release-skill branch 11 times, most recently from 450db0a to d15822f Compare May 15, 2026 02:15
@igorts-git igorts-git force-pushed the add-release-skill branch from d15822f to 7ad7557 Compare May 15, 2026 23:53
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant