HavenSkills

Claude Coding Skills

Build features, investigate defects and make changes easier to review.

Skills
6 skills
Full write-ups
0

How to choose

Start with the smallest behavior change you can verify. Use architecture guidance when the structure itself is limiting the change.

Filter

All skills in this category

tdd

Use short test-first cycles to add or change observable behavior.

by mattpocock
npx skills add mattpocock/skills --skill tdd
Author
mattpocock
Repo
GitHub

diagnosing-bugs

Reproduce a defect, isolate its cause and verify the fix with a regression check.

by mattpocock
npx skills add mattpocock/skills --skill diagnosing-bugs
Author
mattpocock
Repo
GitHub

improve-codebase-architecture

Find modules with tangled responsibilities and plan boundaries that simplify future changes.

by mattpocock
npx skills add mattpocock/skills --skill improve-codebase-architecture
Author
mattpocock
Repo
GitHub

code-review

Review a proposed change for regressions, missing coverage and maintainability issues.

by mattpocock
npx skills add mattpocock/skills --skill code-review
Author
mattpocock
Repo
GitHub

prototype

Build a small prototype to answer a technical or product question.

by mattpocock
npx skills add mattpocock/skills --skill prototype
Author
mattpocock
Repo
GitHub

handoff

Capture decisions, changed files and outstanding tests so another developer can continue the work.

by mattpocock
npx skills add mattpocock/skills --skill handoff
Author
mattpocock
Repo
GitHub

How to evaluate the result

Agent-written code fails in a narrow, predictable way: it does the thing you asked and also changes two files you did not mention. These six skills are arranged against exactly that. tdd keeps the change small enough to see; diagnosing-bugs forces a reproduction before a fix; code-review reads the diff for what you did not ask for; handoff writes down what is still open so the next session does not re-decide it. prototype and improve-codebase-architecture are the two you reach for less often and are the ones most likely to be run too early, because both feel productive while producing very little you can merge. A workable order for a real task is: reproduce, write the failing test, change the smallest thing that makes it pass, review the diff, hand off. Skip the first two and you get a patch that passes on your machine.

How to check the result

Run the tests before you read the diff, not after: a green suite tells you the change did what it claimed, and the diff tells you what else it touched. Then read the diff for edits you did not ask for — renames, formatting, a dependency bump nobody mentioned. That is the failure mode of every agent-written change, and code-review is the skill that catches it if you give it the task description as context. For diagnosing-bugs, keep the failing case as a regression test before you accept the fix; a fix you cannot reproduce the bug against is a guess.

Frequently asked questions

Which coding skill should I start with?

tdd if the behaviour is clear enough to write a test for. diagnosing-bugs if you have a failure but not a cause. improve-codebase-architecture only when the structure itself is what is slowing you down — it is the slowest of the six and the easiest to run too early.

Are these six skills meant to be used together?

They are six skills from one author's engineering set, and they are written to be chained rather than picked from: reproduce the defect, write the test, make the change, review the diff, hand off what is left. Installing one and expecting the rest of the workflow is the usual mistake.

Does code-review replace a human reviewer?

It reviews a proposed change for regressions, missing coverage and maintainability, and returns findings. It does not run your suite — you do that — and it does not approve or merge anything. Treat its output as the review comments of a careful colleague who has read the diff but not your codebase history.

How do I install one of these coding skills?

Same as every skill here: run the install command on the row, or copy the folder to ~/.claude/skills/name/SKILL.md, where name is the skill's folder. The path must end at SKILL.md — an extra folder level on top is the usual reason a skill never activates.