dotkyl

Port recent general tech-writing changes from the work skill (orm-claude-tech-writing) into this repo's copy of the tech-writing skill. Use when the user asks to sync, fold in, or apply recent tech-writing changes from work, update the tech-writing skill from the work one, or catch this repo's skill up to the work one.

Create git commits following the workflow and message conventions. Use whenever the user asks to commit, make a commit, save changes, or any variation of committing work. Always use this skill instead of the default commit behavior.

Guidance for writing technical prose for an engineering audience — PR descriptions, commit message bodies, design docs and RFCs, technical explanations, multi-sentence code comments, incident write-ups, and any explanation of how a system behaves. Use this skill whenever you draft or revise writing whose reader is another engineer, even if the user only says "write up X", "explain this", "draft a PR description", or "write a comment block". Also use it when asked to re-explain or clarify a confusing technical explanation — including phrasings like "I don't understand what this means", "explain this more clearly", "rewrite this so it makes sense", "explain it to me like a robot", "pretend you're a robot and explain again", or "robotify that". These robot-voice and re-explain requests should trigger the skill even when you're restating something you said earlier in the same conversation. This skill governs full-prose writing: for code comments and commit messages it applies to the prose body — a multi-sentence

Add a temporary, self-deleting mise task to root mise.toml that runs a one-time action on the *other* machine after it next pulls, then prints a reminder to delete the task. Use when a change on this machine leaves the other machine needing a one-off action that `mise run sync` / manage-symlinks won't do on its own — e.g. pruning a stale symlink left by a rename, or migrating a config file in place. The action runs once and then the task is deleted, so it's not for durable machine setup. For syncing an app install, use add-install-task instead.

Turn on (or adjust) spoken progress narration via the macOS `say` command, so the user can follow your work by ear while away from the keyboard. The user invokes this with `/say [level]` — they pass a level word as the argument (off, quiet, default, verbose) or nothing for the default level. Narration is off unless invoked. This is user-invoked only; do not trigger it on your own.

Conventions for git commands other than commit. Use whenever you are about to push, amend, or rebase. For commits, use the `commit` skill.

Add a one-time install task to root mise.toml to sync a hand-installed app from this machine to the other one. Use when the user says something like "I installed Vibe Island on this computer, create a task to add it to my other computer". See CLAUDE.md "One-time install lifecycle".

Write or refactor shell scripts in this repo. Use when asked to create a new script in bin/, refactor an existing script, or review a script for style. Applies repo conventions and style principles.

Conventions for writing or editing markdown. Use whenever you are writing or editing a .md or .markdown file.

Sort files from a Google Drive inbox folder into organized destinations based on a rules file. Use this skill whenever the user asks to sort, organize, or file away items in their inbox, mentions "-Inbox", says "sort my inbox", "file these", "organize my downloads", or references their Google Drive Reorg Rules. Also trigger when the user drops files into their inbox and asks Claude to handle them, even if they don't use the word "sort" explicitly.

Show a prioritized dashboard of open PRs that need attention. Covers PRs authored by the user, PRs they've commented on, and keyword-matched PRs across oreillymedia. Use when Katy asks about open PRs, what needs review, what's going on, or wants a PR status check. Also trigger on "my PRs", "what needs my attention", "PR dashboard", or "check PRs".

Guidance on when to use subagents to keep the main conversation context clean. Use this skill when you're about to do research that involves reading many files (5+), exploring unfamiliar parts of a codebase, investigating a bug across multiple modules, or answering a broad question that requires gathering information from scattered sources. Also use when the user asks you to "look into", "investigate", "explore", or "find out how X works" — these are signals that the work is research, not implementation, and a subagent can do it without bloating the main context.