multi-agent-release-manager

name	multi-agent-release-manager
description	Cleans up the workspace, formats code, runs presubmit checks, and uploads CLs to Gerrit.

Multi-Agent Release Manager Protocol

This skill is a single-agent utility designed to handle workspace hygiene, code formatting, final validation, and CL deployment for both Jujutsu (JJ) and Git environments.

Stage 0: Grounding & Validation

1. Discover Environment: Read project.magi.json to ground the VCS (GIT or JJ), repo type (CHROMIUM or GOOGLE_INTERNAL), and temp_directory. If project.magi.json is missing, default to GIT and search for .git or .jj directories to verify.
2. Format Code:
   • Run git cl format (or equivalent formatter for the repository type).
3. Validate Build & Tests:
   • Run git cl presubmit if in Chromium repository.
   • Run the specified build_targets or unit tests defined in project.magi.json to ensure no last-minute formatting bugs broke the build.
   • If tests fail, report the failures and abort the release. Do NOT upload.

Stage 1: Deploy CLs

1. Determine Tag and Hashtag: Read project.magi.json#active_protocol to determine the tagging:
   • MAGI -> Tag: TAG=magi, Hashtag: magi
   • TDD -> Tag: TAG=magi-tdd, Hashtag: magi-tdd
   • REVIEW -> Tag: TAG=magi-review, Hashtag: magi-review
   • If active_protocol is missing, do not add these tags/hashtags.
   • Note: Global rule TAG=agy and CONV=<conversation_id> must always be appended to the description footer.

Based on the VCS detected:

jujutsu (JJ) Workflow

1. Run jj status to identify modified files.
2. Feature CL (Product Changes):
   • Identify all files modified under product source directories.
   • Separate them into their own change (rooted at main). Use jj split or jj squash -i if they are mixed with MAGI files.
   • Describe the change. Append the determined Tag (e.g., TAG=magi) and global tags (TAG=agy) to the commit message footer.
   • Upload the change: jj upload (or equivalent upload command) with the determined Hashtag (e.g., magi).
3. MAGI Upgrades CL (Config Changes - If Applicable):
   • If personas/**/*.json or ROUTING.md were modified (e.g., by training), create a sibling change rooted at main.
   • Stage only the MAGI files in this change.
   • Describe the change. Append the determined Tag (e.g., TAG=magi) and global tags (TAG=agy) to the footer.
   • Upload the change with the determined Hashtag (e.g., magi).

Git Workflow

1. Run git status to identify modified files.
2. Feature CL (Product Changes):
   • Create a clean branch for the feature if not already on one.
   • Stage only the product source changes: git add <product_files>. Do NOT stage any agents/skills/ changes.
   • Commit the changes. Append the determined Tag (e.g., TAG=magi) and global tags (TAG=agy) to the footer of the commit message.
   • Upload the CL: git cl upload --hashtag <hashtag> (using the determined Hashtag).
   • Push description: git cl description -n +.
3. MAGI Upgrades CL (Config Changes - If Applicable):
   • If there are modified files in personas/**/*.json or ROUTING.md, create a separate branch.
   • Stage only the MAGI configuration files: git add <magi_files>.
   • Commit the changes. Append the determined Tag (e.g., TAG=magi) and global tags (TAG=agy) to the footer.
   • Upload the CL: git cl upload --hashtag <hashtag> (using the determined Hashtag).
   • Push description: git cl description -n +.

Stage 2: Workspace Cleanup

1. Revert Leftovers: Revert any untracked or modified files that were not part of the uploaded CLs (e.g., accidental submodule bumps).
2. Delete Temp Files: Find and delete all temporary files generated during the process (e.g., files matching *.magi, *.magi.* in the repository).
3. Delete Temp Directory: Delete the configured temp_directory recursively.
4. Final Status: Report the uploaded CL URLs and confirm that the workspace is clean.

Evaluation & Testing

When modifying this skill's workflow, routing, or schemas, ensure that the corresponding Promptfoo evaluation test suite is updated and passing:

• eval.promptfoo.yaml