Stats
Actions
Tags
From bosskung-plan
Use when implementing multi-step features or tasks that benefit from cost-efficient model tiering — strong planner, fast executor, quality reviewer per task.
How this skill is triggered — by the user, by Claude, or both
Slash command
/bosskung-plan:plan-execute-reviewThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Use three Claude models at their strengths: Opus plans (deep reasoning), Haiku executes (fast/cheap), Sonnet reviews (balanced quality). Never use a heavy model where a lighter one suffices.
Use three Claude models at their strengths: Opus plans (deep reasoning), Haiku executes (fast/cheap), Sonnet reviews (balanced quality). Never use a heavy model where a lighter one suffices.
claude-opus-4-8)Opus produces the implementation plan. Output must be:
Model: claude-opus-4-8
Role: Architect
Output: tasks.md or inline plan
claude-haiku-4-5-20251001)Haiku executes one task at a time from the plan. Feed each task in isolation.
Model: claude-haiku-4-5-20251001
Role: Implementer
Input: single task + acceptance criteria
Output: code diff / file changes
Rules:
claude-sonnet-4-6)Sonnet reviews Haiku's output before moving to next task. Block on failure.
Model: claude-sonnet-4-6
Role: Reviewer
Input: task spec + Haiku's output
Output: PASS or FAIL with specific issues
Review checks:
If FAIL → Haiku retries with Sonnet's feedback. Loop until PASS.
| Phase | Model | When to stop |
|---|---|---|
| Plan | Opus | Plan approved by user |
| Execute | Haiku | Task diff produced |
| Review | Sonnet | PASS verdict |
| Retry | Haiku | After FAIL feedback |
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub thadchet/bosskung-skills --plugin bosskung-skills