skills/step2/SKILL.md
ONLY when user explicitly types /step2. Never auto-trigger on plan, design, or architect.
npx skillsauth add gsa9/3-step-plan-claude-code step2Install 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.
Create _step2_plan.md at repo root. This file is the sole input for /step3 subagents, which have zero prior context.
_step2_plan.md._step2_plan.md is written and _step1_decisions.md is deleted.Read _step1_decisions.md.
If found: transform it, do not copy it. Convert decisions into phase-level tasks. Convert pitfalls and rejected alternatives into guardrails phrased as "Do NOT use X because Y". Embed each ## Constraints entry as a Guardrail in every phase that could violate it.
If not found: use conversation context or ask the user.
Read relevant source files referenced in the decisions document or identified from the goal.
Design all phases, then write the complete _step2_plan.md in one Write call.
Phase design rules:
Do not output anything during these operations.
Delete _step1_decisions.md via Recycle Bin (never permanent delete).
On Windows, delete _step1_decisions.md:
powershell -NoProfile -Command 'Add-Type -AssemblyName Microsoft.VisualBasic; [Microsoft.VisualBasic.FileIO.FileSystem]::DeleteFile((Join-Path "WORKING_DIR" "_step1_decisions.md"),"OnlyErrorDialogs","SendToRecycleBin")'
Verify the file is gone. Show the output box. Stop.
When _step1_decisions.md contains code-task: true, apply these overrides:
**Modifies:** and **Reference:** instead of **Outputs:** and **Inputs:**.Est. Lines column.Use this structure for _step2_plan.md:
---
code-task: true # only if _step1_decisions.md has code-task: true
title: [from _step1_decisions.md frontmatter]
---
<!-- @plan: /step2 -->
# [Goal Title]
**Goal:** [1-2 sentences]
## Rationale
**Approach:** [chosen approach]
**Why:** [key reasons from evidence]
**Patterns:** [conventions subagents must follow]
**Constraints:** [what must hold true and why — from `## Constraints` in decisions]
## Phases Overview
| Phase | Name | Depends | Parallel Group |
|-------|------|---------|----------------|
| 1 | ... | - | A |
| 2 | ... | 1 (produces X) | B |
## Phase Details
### Phase N: [title]
**Outputs:** [exact deliverable names/paths]
**Inputs:**
- `source/or/reference` — [what to look at and why]
**Guardrails:**
- [constraint from `## Constraints` that this phase could violate — with WHY]
- [pitfall or rejected alternative phrased as "Do NOT X because Y"]
**Tasks:**
- [ ] Concrete task 1
- [ ] Concrete task 2
Output:
[title]
▰▰▰▰▰▰▰ ▰▰▰▰▰▰▰ ▱▱▱3▱▱▱
testing
ONLY when user explicitly types /step3. Never auto-trigger on execute, run, or implement.
testing
ONLY when user explicitly types /step1. Never auto-trigger on think, scan, consider, or examine.
testing
ONLY trigger on explicit /gc. Never auto-trigger on 'commit', 'push', or similar words.
testing
ONLY when user explicitly types /distill. Never auto-trigger on optimize, shorten, reduce, or trim.