claude-plugins

Address PR review feedback autonomously in a monitor-fix-push loop until automated reviewers approve or have no blockers. Supports Claude reviews (`claude[bot]`) and Codex reviews (`chatgpt-codex-connector`, `chatgpt-codex-connector[bot]`), including Codex inline P1/P2/P3 comments. Default mode (`--non-interactive`) polls or background-watches for new feedback, triages it, fixes valid issues, replies, commits, pushes, and loops. In Codex, use a cheap read-only background watcher agent when the user asks to wait, watch, loop, background, or wake them when feedback lands and no Monitor tool exists. Pass `--interactive` for a human-confirmed single pass. Use whenever the user says 'pr-review', 'pr-feedback', 'handle PR comments', 'fix PR review', 'respond to PR', 'address Codex comments', 'handle Claude review', 'run the PR loop', 'watch for PR feedback', 'background PR review', or asks to check reviewer feedback.

Generate images from text prompts or edit existing images using Google Gemini's Nano Banana Pro model. Use this skill whenever the user wants to create an image, generate a picture, make a visual, create artwork, design graphics, edit a photo, modify an image, update a picture, or add/remove/change elements in an image. Also trigger when the user says 'nano banana', 'generate an image', 'create a picture', 'make me an image', 'edit this image', 'modify this photo', mentions wanting a logo, icon, illustration, diagram rendered as an image, or references image generation or manipulation in any way - even casually like 'can you make a picture of...' or 'what would X look like'.

Run a local multi-agent code review before pushing or creating a PR. Use this skill whenever the user says 'code-review', 'pre-review', 'review my changes', 'review before push', 'check my branch', 'local review', 'pre-PR review', or wants feedback on their current branch changes before submitting a pull request. Also use when the user asks to 'review the diff', 'check for issues before PR', or mentions wanting to catch review feedback early. This skill spawns multiple specialized review agents in parallel to give comprehensive feedback fast.

Create well-formatted git commits using conventional commit format. Use this skill when the user asks to 'commit', 'commit changes', 'make a commit', 'create a commit', 'commit this', 'commit my work', 'write a commit message', 'stage and commit', or otherwise requests help producing a git commit. Also triggers when the user wants to split a large change into multiple atomic commits, or asks for a good commit message for current work. Handles staging, pre-commit checks, multi-commit splitting, and message drafting in conventional commit style.

This skill should be used when the user asks to "find my open action items", "show my tasks", "what do I need to do", "list my todos", "find tasks assigned to me", "find tasks for [name]", or "what are my open items". Retrieves all open action items from meeting notes (tasks tagged with a person's name) and daily notes (unchecked checkboxes) across the Obsidian vault.

Operate Obsidian vaults via the `obsidian` CLI - create/read/edit notes, manage properties and tags, search vault content, handle daily notes and templates, manage plugins and themes, work with bases, control sync and file history, manipulate workspace layout, and use developer tools. Use when the user asks to interact with their Obsidian vault, manage notes, query vault data, automate Obsidian workflows, or debug Obsidian plugins. Triggers: mentions of Obsidian, vault operations, note management, obsidian CLI, daily notes, or working with .md files in an Obsidian context.

Help users create, edit, debug, and optimize Cassette meeting transcript processing flows. Cassette is a CLI tool that watches for transcript files (JSON/VTT), runs them through LLM prompt chains, and outputs structured Markdown. Use this skill when the user mentions cassette, meeting transcripts, transcript processing, prompt chains for meetings, or wants to set up automated meeting note generation. Also use when the user wants to generate Obsidian-formatted meeting notes, debug why a transcript failed processing, write or improve cassette prompts, or design multi-step flows for transcript cleanup and summarization.

Find and rank top YouTube videos for any topic. Searches YouTube via yt-dlp with query expansion and scores results by views, engagement, recency, and channel authority. Use when the user asks to find top/best YouTube videos, search YouTube, rank videos, or discover quality video content on a topic.

Find one insight that eliminates multiple components - "if this is true, we don't need X, Y, or Z"

Use when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation