Claude code workflow skills for planning, TDD and debugging

Claude code workflow skills change how an agent works through a task: ask questions first, write a plan, test before coding, debug in a fixed order, hand work to a fresh session. This is our largest category, with 104 skills and packs reviewed. 53 get an install verdict and 51 come with caveats.

Size makes choosing harder. Many packs overlap, and installing several at once gives the agent competing instructions about how to start a task. Pick one primary workflow, then add small single purpose skills around it.

What to look for in Claude code workflow skills

Check how much the skill interrupts. Interview skills such as grilling wait for your answers every round, which is the point, but it rules them out for unattended runs. Check what it writes: plan files, glossaries, issues on your tracker. A skill that creates and closes issues needs your tracker configured first.

Then check automation. Hooks, which Claude Code runs on events, are common in this category, and so is pre-approved shell access. A hook that fires on every prompt and tool call is fine if you chose it and want it. Our status labels flag these cases: of the 104 skills, 20 are clean, 59 are marked “review” and 25 have broad access. The security page has the full flag list.

The Claude Code docs say a skill’s body stays in the conversation once loaded, so long process packs cost context for the whole session. Prefer the shorter skill when two do the same job.

Recommended workflow skills

• Matt Pocock’s Skills is a set of 38 small skills (specs, tickets, TDD, code review, handoff) with about 30.5 million recorded installs across the set. The skills average 4.5 KB of Markdown, and nothing makes network calls on its own. Depth per skill is modest, and you can take them one at a time.
• Superpowers is the fuller methodology: 15 skills covering brainstorming, TDD, planning, review, debugging and git worktrees. A session hook loads its introduction skill every time. It is a poor fit for one line fixes.
• Agent Skills by Addy Osmani offers 25 lifecycle skills and 9 slash commands. No skill declares allowed-tools, and the optional hooks are shell scripts you wire in yourself.
• handoff is clean and tiny (1.5 KB). It writes a summary file to your temp folder so a fresh session can continue. The agent redacts secrets, so check the file before sharing it.
• Planning with Files keeps task_plan.md, findings.md and progress.md on disk and re-injects them through hooks, so a plan survives /clear and compaction. We mark it broad access because it pre-approves unrestricted Bash and runs hooks on every prompt and tool call.

A last caution on wayfinder. It plans large work as a map of decision tickets on your issue tracker, and it creates, assigns, comments on and closes issues. That earns a broad access label even though the skill is Markdown only. Set up your tracker first and test it on a scratch repository.

If you only need review, see code review skills. For context between sessions, see memory skills. The full table lists all 104, and how we review explains depth scores.