dotfiles

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

Collaboratively build or edit strong LLM instruction files — skills (SKILL.md), agent instruction files, and slash-command instructions — by orchestrating three specialist skills: skill-creator (mechanics + authoring), prompt-engineer (LLM-instruction quality), and doc-coauthoring (prose, structure, economy). Use this whenever the user wants to create a new skill or agent file, workshop a draft, or edit/improve an existing one and wants it reviewed for quality before landing — even if they only say "make a skill for X", "tighten this skill", "fix the description on my agent", or "help me write a SKILL.md". Prefer this over invoking skill-creator, prompt-engineer, or doc-coauthoring alone when the goal is a production-quality instruction file.

Scaffold or refresh a project- and branch-specific session memory file that captures turn protocol, current task list, decisions log, and open ambiguities for the current git repo. The file is written to the session's auto-memory directory and indexed in MEMORY.md so it auto-loads on future sessions in the same CWD on the same branch. Trigger when the user types `/project-init`, asks to "set up this project/branch", "kick off work on this repo", "init project memory", "scaffold this workspace", "create a session memory file for this branch", or otherwise asks for branch-scoped persistent memory and turn protocol. Trigger eagerly when the user is starting fresh work in a repo and wants per-branch memory across sessions, even without naming the skill explicitly. Do NOT trigger on session open or on a fresh session by itself — this skill is opt-in.

Code review best practices. Use when user wants to review code, asks for code review checklist, or mentions "code review."

Autonomously generates and executes a conventional commit. Use when the user says: "commit", "make a commit", "conventional commit", "/commit", or "commit my changes". Inspects staged/unstaged changes, derives the correct type and scope, writes a Conventional Commits-compliant message, and runs `git commit` without the user needing to write the message manually.

Create a README.md file for the project. Use when the user asks to generate, write, or update a project README.

Guide users through a structured workflow for co-authoring documentation. Use when user wants to write documentation, proposals, technical specs, decision docs, or similar structured content. This workflow helps users efficiently transfer context, refine content through iteration, and verify the doc works for readers. Trigger when user mentions writing docs, creating proposals, drafting specs, or similar documentation tasks.

Resolves environment variables, including tool-agnostic abstractions like $AI_CONFIG_DIR and $MEMORY_DIR. Use when any skill or agent needs to look up an environment variable or resolve a path from the environment.

Generates, creates, and maintains a Pull Request title and description based on branch commits. Use when the user says: "create PR", "open a PR", "generate PR description", "write a PR", "update PR description", "review the PR", "PR title and description", or "update the PR". PR title follows Conventional Commits format. Updates an existing PR when new commits are added or a review is requested.

Expert at writing, reviewing, and improving prompts and instruction files for LLMs (Claude, GitHub Copilot, ChatGPT, or any AI assistant). Use this skill whenever the user wants to write a new system prompt, CLAUDE.md, agent instruction file, copilot-instructions.md, skill file, or any directive intended for an AI — or when they want to improve, audit, or debug an existing one. Also triggers when the user says their AI "isn't following instructions", "keeps hallucinating", "doesn't understand the task", or asks why a prompt "isn't working". The goal is always the same: zero ambiguity, 100% task confidence.

Test-driven development with red-green-refactor loop. Use when user wants to build features or fix bugs using TDD, mentions "red-green-refactor", wants integration tests, or asks for test-first development.