Conventions
Every skill in this hub follows the same conventions. Once you’ve used one, the rest feel familiar.
AskUserQuestion, not plain text
Section titled “AskUserQuestion, not plain text”All user confirmations go through the AskUserQuestion tool — never a plain-text “Proceed? (y/n)” prompt. Why: plain-text prompts are easy to miss, don’t offer structured options, and don’t integrate with Claude Code’s UI. AskUserQuestion gives you a clean multiple-choice UI with an “Other” escape hatch for free input.
Practical effect: if a skill is about to do something irreversible (commit, push, .env overwrite, 1Password item creation), you’ll see a structured prompt with explicit options. Approvals are per-step, not per-run.
Identity verification before writes
Section titled “Identity verification before writes”Skills that touch cloud accounts (bedrock-ops) verify the active identity before any write operation. If sts get-caller-identity doesn’t match the account ID in aws-project.json, the skill aborts rather than writing to the wrong account.
This matters more than it sounds. “Oh, wrong profile” is a common mistake, and cloud mistakes are rarely free to undo.
Dry-run first, execute second
Section titled “Dry-run first, execute second”S3 sync, S3 cp, and similar write operations always dry-run first and require an explicit --execute to actually mutate. You see what would change, confirm, then the real run happens. No silent destructive operations.
Language choice per run
Section titled “Language choice per run”Skills that produce written output (release notes, cost reports, diagram companion guides, DevelopersIO articles) let you pick the language at the start of the run. Options are usually English, Japanese, and “Other” for free input.
Commit messages are the exception — they always stay in English regardless of your language choice, following git convention.
Config files for repeat preferences
Section titled “Config files for repeat preferences”The skills that have a lot of prompts on first run (release, bedrock-ops) offer to save preferences to a project-local config.json at the end of the first successful run. On subsequent runs, the prompts are skipped and the saved values are used.
You can always override by editing or deleting the config.json.
Secrets never in Claude’s context
Section titled “Secrets never in Claude’s context”Skills that touch API tokens (devio with Contentful’s CMA token, migrate-env-to-1password with 1Password credentials) use a helper script that reads the secret directly from the environment or .env. The token never enters Claude’s context, so it can’t accidentally end up in a log, a commit, or a paste.
Frontmatter: user-invocable
Section titled “Frontmatter: user-invocable”Skills with user-invocable: true in their SKILL.md frontmatter can be called as /skill-name slash commands in Claude Code. Not all skills are user-invocable — some are meant to be triggered implicitly when Claude detects a matching intent. Each skill’s page notes whether it’s user-invocable.
Japanese-flavored skills
Section titled “Japanese-flavored skills”Two skills — migrate-env-to-1password and devio — have Japanese-heavy prompts and output. They’re marked with a “Japanese” sidebar badge on the skill catalog. You can use them from any locale of this hub, but expect 日本語 in the interactive flow.
If that’s wrong for your team, the workaround is to fork and translate the prompts.