Use at the start of a complex or long task - anything past ~5 tool calls, multi-service/multi-environment work, production writes, debugging with several hypotheses, cross-repo refactors, or resuming work across sessi…
---
name: task-kickoff
description: Use at the start of a complex or long task — anything past ~5 tool calls, multi-service/multi-environment work, production writes, debugging with several hypotheses, cross-repo refactors, or resuming work across sessions. Sets up the session right: plan/implement separation, externalized state, human-in-the-loop only at high-leverage gates, stop-after-two-failures, and multi-level verification. Skip for single lookups or pure chat.
---
# task-kickoff
The opening protocol for human+AI collaboration on a long task. Most agent failures
happen in the first few minutes — planning and implementing at the same time, never
externalizing state, stutter-stepping the human on every action, or grinding the
same broken approach a third time. This skill front-loads the decisions that prevent
those.
## When to use
Fire if **any** holds:
- The user says "kickoff" / "let's start" / "spin this up" (or your own kickoff
trigger).
- The task will take more than ~5 tool calls.
- It touches **production**, multiple clusters/environments, multiple services, or
any write (deploy, apply, push, DB/config mutation).
- It's debugging/troubleshooting with several independent hypotheses to test.
- It's a cross-repo refactor.
- It resumes earlier work ("pick up yesterday's plan").… install to load the full skill