Kyle Kestell's personal Claude Code / Codex skill library.
npx claudepluginhub kkestell/skillsA structured workflow for longer-lived coding work: init, start-task, research, brainstorm, plan, work, verify, end-task.
k is a structured workflow plugin for longer-lived coding work. It gives Claude a task lifecycle (init, start-task, brainstorm, plan, work, verify, end-task) plus supporting skills for research, commit hygiene, doc review, and language-specific guidance.
Task state lives under .k/, including:
.k/current_task.json.k/tasks/<task_id>/brainstorm.md.k/tasks/<task_id>/plan.md.k/tasks/<task_id>/research/*.mdclaude plugin marketplace add kkestell/skills
claude plugin install k@kkestell
Update:
claude plugin marketplace update kkestell
claude plugin update k@kkestell
Uninstall:
claude plugin uninstall k@kkestell
claude plugin marketplace remove kkestell
Clone the repo, then add it as a local marketplace:
git clone https://github.com/kkestell/skills.git
From inside Claude Code:
/plugin marketplace add /path/to/skills
Then install k from the plugin UI. The marketplace points at your local directory, so edits to skill files are picked up on restart or /reload-plugins.
| Skill | What it does |
|---|---|
init | Initialize .k/ directory (once per repo) |
start-task | Create a task branch and task context for new work |
brainstorm | Explore a fuzzy problem before committing to a plan |
research | Research one topic and save a reusable task note |
plan | Write a concrete plan from an idea, bug report, or brainstorm |
work | Execute the plan on the active task branch |
verify | Check the work against a supplied plan or spec |
end-task | Merge the task branch back to main |
| Skill | What it does |
|---|---|
commit | Clean, deliberate commits |
deslop | Audit user-facing docs for AI slop |
go-lang | Go-specific guidelines (required before editing .go files) |
rust-lang | Rust-specific guidelines (required before editing .rs files) |
Run init once per repo, then use start-task to choose whether the task should run in a sibling worktree or in-place on a task branch. brainstorm is optional; research can be used from brainstorm, plan, or work whenever the task needs evidence.
verify does not require active task state; point it at the plan or spec you want to check against.
Claude currently exposes these as plain slash commands like /init and /start-task, with (k) shown in the picker. /k:init is not a valid invocation.
/init
/start-task <topic> # asks whether to create a worktree
# -- if you chose a worktree, continue in the new Claude terminal session there --
/brainstorm <description>
/research <topic> # optional, repeat during brainstorm/plan/work
/plan <@brainstorm | description>
/work <@plan>
/verify <@plan | path/to/spec.md>
/end-task
Use /commit during /work whenever a tested logical chunk is ready to checkpoint.
start-task can create either a sibling worktree or an in-place task branch.xdg-terminal-exec, then x-terminal-emulator, then xterm.cd <worktree> && claude instruction.end-task retains a separate worktree after merging so post-merge validation can still run cleanly on main./commit, /deslop, /go-lang, and /rust-lang are usually invoked from /work or /verify, not used as the main workflow spine.Open Design — local-first design app exposed to coding agents over MCP. Install once with your agent's plugin command and projects/files/skills are reachable through stdio.
Claude Code plugins for the Slidev presentation framework
Bundled plugins for actuating and debugging the Chrome browser.