llm-tooling

Decides what in the working tree belongs in a commit, writes a Conventional Commits message, and creates the commit locally. Invoked as /llm-tooling:commit, or when the user asks to "commit this", "commit my changes", or "write a commit message". Excludes review and planning artifacts, and asks when the right contents are unclear. Never pushes.

Interactively builds a "Gauntlet Loop" prompt — a goal-plus-quality-bar prompt that a coding agent runs as a builder/critic loop until the output matches or beats a reference standard. Use when the user says they want to "create a gauntlet loop", "run a gauntlet loop", "write a gauntlet prompt", "set the bar for a feature", or asks for help turning a vague feature idea into a requirements-driven agent prompt. Works for greenfield products and for changes to an existing codebase, with or without a reference implementation. Do NOT use to actually implement the feature — this skill produces the prompt, not the code.

Creates a reveal.js HTML presentation from markdown content, a topic description, or rough notes. Use when the user asks to "create a presentation", "make slides", "build a slide deck", "create a reveal.js presentation", or provides content and asks for a presentation. Accepts optional theme preference and output path. Do NOT use for PowerPoint, Google Slides, Keynote, or W3C Slidy formats.

End-to-end path from a change request to implemented, verified code: reads the request from wherever it lives (a markdown file, a Jira ticket, a Notion page, a GitHub issue, or the prompt itself), runs the implementation-planner agent to draft an approach, runs it past the implementation-plan-reviewer agent for unnecessary complexity and missed detail, develops it into an agreed plan with the user, implements it, verifies it with the implementation-reviewer agent (which owns running the tests and linters), and offers both to open a pull request and to archive the plan and outcome to a Notion database of implementation artifacts. Invoked as /llm-tooling:implement-change, or when the user asks to "implement this", "build this feature", "work this ticket", "make this change", or "plan and implement".

Writes a reviewer-focused description for the current branch and opens the pull request on GitHub. Invoked as /llm-tooling:pr-create, or when the user asks to "open a PR", "put up a PR", "create a pull request", or "write a PR description". Opens a draft by default, for a self-review with pr-review before colleagues are asked. Accepts draft status, title, base branch, labels, reviewers, and assignees. Uses the GitHub MCP server; falls back to the gh CLI only with the user's explicit approval.

End-to-end handling of review feedback on a pull request: pulls the PR's unresolved comments, runs the github-pr-fix-planner agent to plan a response to each one, walks the user through them, implements the approved fixes, verifies and reviews the result, pushes, and replies on each comment thread. Invoked as /llm-tooling:pr-fix, or when the user asks to "address review comments", "fix the PR feedback", or "respond to my reviewer". Uses the GitHub MCP server; falls back to the gh CLI only with the user's explicit approval.

Pull request review: runs the github-pr-reviewer agent on a PR and reports its major weaknesses; on request, goes on to draft review comments, iterate on them with the user, and post the review to GitHub. Invoked as /llm-tooling:pr-review, or when the user asks to "review this PR", "what's wrong with this PR", "review and post comments", "submit a PR review", or "publish my review". Uses the GitHub MCP server; falls back to the gh CLI only with the user's explicit approval.