How to use the Antigravity Maintainer Batch Release agent skill

What is an agent skill, and what does this one do?

An agent skill is a folder containing a SKILL.md file and supporting reference files. When an AI coding agent such as Claude loads the skill, it reads these instructions and follows them for matching tasks.

This skill tells the agent how to run maintainer work on the agentic-awesome-skills repository: triaging and reviewing pull requests in batches, running protected merge commands, preparing and publishing releases, and verifying public catalog and Pages surfaces. It names the checks, scripts and reference documents to follow.

• Runs maintainer PR sweeps and merge batches with protected main and admin enforcement kept on.
• Prepares and publishes releases to npm and GitHub, and checks the published bytes.
• Verifies catalog, Pages and legacy redirect changes.
• Covers Core checks for CLI, MCP and Workbench changes.
• Does not cover ordinary contribution work such as submitting a skill or fixing a bug.

Check that you are allowed to use it

The skill is written for maintainers. Its merge and release steps need repository and publishing rights. Without them, the skill is not useful and misuse could cause public changes.

The skill says ordinary contributors do not need maintainer privileges, and it does not cover their work.

• Node.js and npm, to run the repository scripts such as npm run merge:batch, maintenance:cycle, release:prepare and release:publish.
• Git and GitHub access with maintainer permissions on the repository.
• A private state directory outside source checkouts, with mode 0700.
• npm publishing credentials and GitHub release rights, for release steps only.
• Optional: Tessl review access. Without it, the skill records manual-review-required rather than a pass.

Install the skill

Download the ZIP for this skill from this page, then choose one of the following options.

For a personal install, unzip it into ~/.claude/skills/. For a project-specific install, unzip it into the project's .claude/skills/ folder. You can also upload the skill in the Claude apps under Settings → Capabilities → Skills.

• Personal: unzip into ~/.claude/skills/
• Project: unzip into .claude/skills/ inside the project
• Claude apps: Settings → Capabilities → Skills, then upload

Trigger the skill with natural requests

You do not need a special command. Describe the maintainer task in plain language, and name the repository and the work involved. The skill's description matches requests about repository maintenance, main alignment, CLI, MCP or Workbench changes, and release work.

Requests that fit well include a batch of named pull requests, a release to prepare, or a check of public surfaces after a change. Requests about submitting your own skill are outside its scope.

• Review PRs 123 and 124 in the agentic-awesome-skills repo and merge them in a batch if they pass.
• Prepare release 1.2.3 of the runtime, following the maintainer release steps.
• Verify the catalog and Pages surfaces after this catalog change.

Run a safe analysis before any merge

The skill's example shows how to prepare analyses without merging. Use --prepare-only first so you can review the evidence.

Merging requires a real semantic review of the complete bundle. After that review, the skill persists the evidence in a private state directory and merges using the exact full commit SHA. Use real IDs, paths and a full immutable SHA, and make sure the state directory already exists, is privately owned, has mode 0700, and sits outside source checkouts.

Safety rules the skill enforces

The skill sets several hard limits. It never pushes directly to main, never bypasses protections, and never uses GitHub auto-merge. It does not treat a heuristic score as a substitute for semantic review. It does not rewrite PR metadata or automatically retry base drift.

The skill runs scripts and uses the network. Read the scripts it runs before using it on a real repository, and run merge and release steps only with the credentials and authority you actually hold. Its own limitations section says it grants no additional publication or messaging authority, and that SkillSpector and timing reports are advisory only.

• Never push directly to main or use admin bypass.
• Do not use mutable artifact identity or PR-controlled review receipts.
• Tessl unavailable means manual-review-required, not a passing review.
• Closing a PR requires a posted owner rationale bound to its exact head, through maintenance:cycle close.

Tips for using it well

Start with analysis rather than merges, and keep state in a private directory outside your checkout. Work through the references the skill points to, since they hold the exact validation profiles and release proof. Treat the skill as a checklist to follow with your own judgement, not as an autonomous release button.

• Read the maintenance contract and component validation references before the first real batch.
• Run maintenance:cycle verify on exact current main before preparing a release.
• Keep the repository's protections on; the skill assumes they are enforced.
• Report personal MCP configuration and session reload separately, as the skill instructs.