agents/dot-agents/skills/execute/SKILL.md
Look at the todo list, choose the next most appropriate actionable todo, execute exactly one todo, then stop and report back on what was done. After the work, reconcile the todo list so related todos reflect reality. Use this whenever the user wants you to pick the next task from todos and make progress without manually selecting one first.
npx skillsauth add nathankoerschner/dotfiles executeInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Choose one todo, do it, and stop.
The point of this skill is to convert a backlog into the single best next action without trying to clear the whole queue. Be decisive, but stay grounded in the actual repo state and the todo metadata.
Treat $ARGUMENTS as optional guidance that can influence selection, for example:
If no arguments are provided, infer the next best todo from the todo list and current project context.
$ARGUMENTSAfter executing the chosen todo, do a short wrap-up pass on the todo system before stopping.
This is a todo reconciliation pass, not permission to execute more work. You are still only allowed to perform one todo's implementation work. The goal is to leave the todo list cleaner, more truthful, and better sequenced based on what just happened.
After finishing the implementation work and reconciling todos, create a git commit before giving the final response.
Do not use the commit as an excuse to bundle unrelated work. The commit should reflect exactly the one todo you executed.
Stop after this one todo and report:
Do not continue to another todo unless the user explicitly asks you to run /execute again.
development
Deep, rigorous multi-model code review for PRs, branches, diffs, commits, or pre-merge checks. Use this when the user explicitly asks for a deep review, thorough review, rigorous review, pre-merge check, PR review, branch review, or wants maximum confidence. Do not use for quick/casual reviews; use quick-review instead. Treat AI output as a first draft: run independent Pi subagents on Claude Opus, GPT-5.5 extra-high reasoning, and Gemini; validate and synthesize severity-ranked findings before fixes.
development
Review UI code for Vercel Web Interface Guidelines compliance
development
Opinionated constraints for building better interfaces with agents.
databases
Connect to Snowflake via SnowSQL