ruicsh/ruicsh-skills
Personal agent skills for routing work to the right workflow, planning, implementing plans task-by-task, committing work, reviewing changes, and gating a release.
Implements a task plan one task at a time under test-driven discipline, with one commit per task. Use when the user says "build", "build auto", "implement the next task", or "work through the plan". Use when a plan or spec already exists and the work must land as verified, individually committed slices.
Assists with committing the working tree into logical Conventional Commits. Use when the user wants to commit changes, split a working tree into clean commits, or write a commit message.
Writes the smallest complete change — reuse what exists before adding code, scope the blast radius first, keep a test for real logic, never cut safety. Always on once loaded. Use on any coding task and when asked for the simplest or YAGNI solution.
Turns a spec into tasks/plan.md — ordered tasks with acceptance criteria, sized for the build skill. Use when the user says "plan", "make a plan", "break this into tasks", or "what are the steps". Use before implementation when the work spans more than one file or session. Writes the plan only; never implements.
Reviews code changes across correctness, readability, architecture, security, and performance. Use when asked to review a diff or PR without modifying files.
Runs the pre-launch gate — parallel review, security, and test passes, then one go/no-go decision with a rollback plan. Use when the user says "ship", "release", "deploy", or "is this safe to launch", or asks for a pre-launch checklist verdict. Reports a decision; never deploys.
Routes a request to the matching skill. Use at the start of a task, when the work changes shape, or when asked which skill or workflow applies. Use when requirements are unclear, when a change crosses planning, implementation, review, or release stages, or when you need to decide between similar skills.
Keeps replies terse and fact-dense — answer first, style compressed, every technical fact kept. Use when the user asks for terse, blunt, short, or no-fluff replies, or when output must cut filler without losing code, paths, numbers, or negations.