Claude Skill MartBrowse skillsWhat is a Skill?
 
← Back to list

Pre-Publish Review (16-Agent Release Gate)

A release gate skill that fans out 16 parallel agents to audit every unpublished change and returns a SAFE/CAUTION/RISKY/BLOCK verdict before you npm publish.

Dev & CodingAdvanced★ 67,531⑂ 5,507AI score 6/10Last updated: Aug 9, 2026

What it does

Detects everything that changed since the last npm release, reviews it through three agent layers, and rules on whether the release is safe to ship.

  • Phase 0: runs /get-unpublished-changes to diff the published version against local, collecting commits, diffs, and changed files
  • Phase 1: splits the changeset into up to 10 logical groups by module/feature area
  • Phase 2: launches, all in one turn and all in background, up to 10 ultrabrain per-group deep-dive agents + a /review-work coordinator (5 agents) + 1 oracle for release synthesis
  • Phase 3–4: aggregates results into a SAFE / CAUTION / RISKY / BLOCK verdict, per-layer semver recommendations, an exhaustive breaking-change list, a changelog draft, and post-publish monitoring advice

Every deep-dive agent works from a 10-point checklist: intent clarity, correctness, breaking changes, pattern adherence, edge cases, error handling, type safety, test coverage, side effects, and release risk.

Note: by design a plain "publish this" request must NOT trigger it — the skill only fires when you explicitly ask for a pre-publish review.

Who it's for

  • npm library maintainers who ship often and fear regressions or accidental breaking changes
  • Monorepo teams that must decide semver separately for core packages and adapters
  • Developers who want changelogs and release notes drafted automatically
  • Advanced users studying multi-agent orchestration patterns

Example uses

  1. Final gate before a release — you ask for a pre-publish review; it splits 37 commits since v1.4.2 into 10 groups, surfaces 2 CRITICAL findings, and returns BLOCK before anything ships.
  2. Version bump decisions — it spots a new required field in the config schema and recommends MAJOR instead of PATCH, complete with user migration steps.
  3. Automatic changelog draft — you get a release note grouped by feat/fix/refactor/breaking/docs, ready to paste into a GitHub Release.

· · · Install guide · · ·

Install in the Claude app (no terminal)
  1. Download the ZIP with the button below.
  2. In Claude, open Settings → Capabilities and turn on 'Code execution and file creation'. (one time)
  3. Go to Customize → Skills → + → 'Upload a skill' and upload the ZIP.
↓ Download ZIP
Install in Claude Code

Let Claude do it — paste this into Claude Code

Install the skill I found on Claude Skill Mart.
Copy the .opencode/skills/pre-publish-review folder from the GitHub repo code-yeongyu/oh-my-openagent into my ~/.claude/skills/code-yeongyu-pre-publish-review/.
When it's done, tell me in one line what this skill can do.

Install with a command instead

git clone https://github.com/code-yeongyu/oh-my-openagent.git /tmp/oh-my-openagent && mkdir -p ~/.claude/skills/pre-publish-review && cp -r /tmp/oh-my-openagent/.opencode/skills/pre-publish-review/* ~/.claude/skills/pre-publish-review/

⚠ This is a third-party skill. Check the source repository before installing.

  1. Open a terminal and clone the repository into a temp folder: 
    git clone https://github.com/code-yeongyu/oh-my-openagent.git /tmp/oh-my-openagent
  2. Create the Claude Code skills directory: 
    mkdir -p ~/.claude/skills/pre-publish-review
  3. Copy the skill files in: 
    cp -r /tmp/oh-my-openagent/.opencode/skills/pre-publish-review/* ~/.claude/skills/pre-publish-review/
  4. Open ~/.claude/skills/pre-publish-review/SKILL.md and adapt the repo-specific dependencies: ultrabrain, oracle, /review-work, /get-unpublished-changes. Replace them with standard subagents or your own prompts if you don't have them.
  5. Replace the hardcoded oh-my-opencode package name and the three "release layer" names with your own project's structure.
  6. Restart Claude Code, open the repo you intend to publish, and say "run a pre-publish review".
  7. The first run is token- and time-heavy because of the agent fan-out — try it on a small changeset first.
View source on GitHub ↗License: NOASSERTION