pick-the-right-skill

Pick the right skill

A collection nobody can navigate is a collection nobody uses.

Grouped by the sentence you would say out loud, not by category. Find your sentence, take the skill.

I am about to change something

• route-work-to-the-right-model — Use before starting any change, to decide how much validation it needs and which model and effort to spend on it. Answers "is this a one-review change or a three-review change?" without arguing about it.
• write-the-rollback-plan-first — Use before applying any change that can break something. The way back is written and tested before the change, not improvised after it.
• deploy-to-production-safely — A change is about to touch a running system. Backup, dry run, rollback and per-node verification, in that order.
• edit-a-live-config-safely — Editing a configuration file a running process is reading. Unique anchor, keep the inode, validate before applying, verify every node.
• check-if-data-is-safe-to-delete — Deciding whether a table, a file, or a service can be deleted because it looks unused. Measure real access, rename before deleting, and check the units of every counter.
• block-dangerous-git-commands — Set up Claude Code hooks to block dangerous git commands (push, reset --hard, clean, branch -D, etc.) before they execute. Use when user wants to prevent destructive git operations, add git safety hooks, or block git push/reset in Claude Code.
• know-what-a-change-costs — Shipping something that runs repeatedly — a job, a query, an agent loop, a pipeline

Something is broken

• diagnose-a-hard-bug — Diagnosis loop for hard bugs and performance regressions. Use when the user says "diagnose"/"debug this", or reports something broken/throwing/failing/slow.
• root-cause-analysis-first — Something is failing and you are about to change code or config to fix it. Blocks the fix until you can say why it happens, and shows how to hunt a cause instead of confirming a guess.
• debug-a-silent-failure — Something did not happen and nothing complained — no error, no alert, an empty log, a change with no effect. Teaches how to read absence as a symptom and where silence gets manufactured.
• triage-issues-and-prs — Move issues and external PRs through a state machine of triage-issues-and-prs roles, categorise, verify, grill if needed, and write agent-ready briefs.
• survive-someone-elses-breaking-change — Something that worked yesterday stopped working and you changed nothing, or before upgrading a dependency, runtime or platform
• make-the-next-failure-loud — An incident, or before shipping something whose failure would be invisible