pwguler-skills

Shape a codebase's architecture, at the start or once it fights you. Use when setting up a new codebase or area and choosing its structure and conventions, or when the user wants the architecture improved: refactoring targets, modules too entangled to change apart, code that is hard to test or hard for an agent to find its way around.

Internal support skill, the relentless interview loop other skills delegate to. Use only when another skill or the user explicitly triggers it, never on your own.

Find the root cause of a bug before fixing it. Use whenever the user says "fix", "broken", "not working", "error", "failing", or "bug", and whenever a bug, test failure, or unexpected behavior appears, before proposing any fix.

Full ceremony mode for the whole suite. Run /deep before risky or ambiguous work; /deep off returns to fast-first.

Question a plan or design until every decision in it is settled and agreed. Use when the user wants a plan stress-tested or a design pinned down before building, or says "drill this" or "drill me".

Implement a settled plan test-first, smallest slice at a time. Use when starting implementation of a feature or fix after the plan is settled, or when user says "implement this" or "build it".

Finish a development branch. Use when implementation is complete and the work needs to be merged, pushed as a PR, kept, or discarded.

Dispatch independent tasks to concurrent subagents. Use when two or more failures or tasks are independent, sharing no files or state, and can be worked simultaneously.

Build a throwaway UI prototype to react to before committing. Use when a design fork turns on look, feel, or flow that prose cannot settle, or when the user asks to prototype or mock something up. Prefers the /design canvas; falls back to a self-contained HTML artifact when /design is not available.

Test your understanding of code the agent wrote. Run /recall after a slice you did not write, or when onboarding into an area. The agent asks, you answer; it never explains first.

Find out what is true about a library, service, or standard by reading whoever owns the facts, and write the answer down with a citation per claim. Use when the user asks to look something up, to check how an API or tool really behaves, or to hand off reading while other work continues.

Judge work whose quality no command can settle, such as "this API design is good" or "this UI is not slop", against criteria written down before the work is looked at. Use when the user asks how something looks or whether it is any good, when a UI, public interface, name, or doc needs a quality call before it ships, when a spec criterion names a quality rather than a behavior, and when verify meets a claim its commands cannot settle.

Report a finished task as its result, not its transcript. Use when a task that produced an artifact or a run (a build, a workflow, a background agent, an implement or verify run, a release) has just finished and you are about to report it, and when the user types /tldr.

Prove a claim of done, fixed, or passing by running the commands and reading the output before making it. Use before claiming work is complete, fixed, or tested, and before committing or opening a PR.

Create or edit a document an agent reads and prove it changes behavior. Use when writing or editing a skill, AGENTS.md, CLAUDE.md, a system prompt, or other agent-facing prose, or verifying one before it ships.