autoplan

name	autoplan
description	Auto full review pipeline. Runs office-hours → CEO review → Eng review → Design review → DX review in sequence, auto-deciding where safe and asking where needed. Use when you want to "review everything", "run the full pipeline", or before major implementation work.
allowed-tools	bash, read, write, edit, grep, find
compatibility	requires git

AutoPlan — Full Review Pipeline

Run the complete planning and review pipeline in sequence. Auto-decide safe choices, escalate taste decisions and safety concerns.

Decision Principles

1. Auto-decide when both models agree — if the recommended choice is clear, proceed without asking.
2. Ask for taste decisions — design aesthetics, naming, UX preferences need human judgment.
3. Escalate safety issues — data loss, security vulnerabilities, destructive scope changes always require confirmation.
4. Stop at gates — if CEO review says HOLD, don't proceed. If Eng review flags architecture issues, fix before continuing.
5. Track decisions — every auto-decision and user choice recorded for audit.
6. Taste ≠ Risk — don't escalate preferences as safety issues, and don't auto-decide security concerns.

Brain Context Load

Before running the pipeline, search your brain for planning context:

1. Extract keywords from the feature and codebase area.
2. Use gbrain_search to find past autoplan runs, design docs, or architecture decisions.
3. Use gbrain_get to read the top 3 matches.
4. Use this context to avoid re-exploring already-decided questions.

If gbrain tools are not available or return no results, proceed without brain context.

---

Pipeline Sequence

Run each skill in order. Each phase consumes the output of the previous phase.

Phase 0: Intake

BRANCH=$(git branch --show-current)
echo "BRANCH: $BRANCH"

Read existing plan if any:

REMOTE=$(git remote get-url origin 2>/dev/null | sed 's|.*[:/]\([^/]*/[^/]*\)\.git$|\1|;s|.*[:/]\([^/]*/[^/]*\)$|\1|' | tr '/' '-' | tr -cd 'a-zA-Z0-9._-')
SLUG="${REMOTE:-$(basename "$PWD" | tr -cd 'a-zA-Z0-9._-')}"
ls -t ~/.gstack/projects/$SLUG/*-design-*.md ~/.gstack/projects/$SLUG/ceo-plans/*.md 2>/dev/null | head -3

If no plan exists, start from office-hours. If a plan exists, ask whether to start from the beginning or from a specific phase.

Phase 1: Office Hours (if no design doc exists)

Execute /skill:office-hours methodology:

• 6 forcing questions in Startup mode
• Produce design document at ~/.gstack/projects/$SLUG/
• STOP if office-hours reveals the idea is not viable

Phase 2: CEO Review

Execute /skill:plan-ceo-review methodology against the design doc:

• Nuclear scope challenge (Expansion, Selective, Hold, Reduction)
• Architecture review, error handling, deployment, observability
• Produce CEO plan at ~/.gstack/projects/$SLUG/ceo-plans/
• Auto-decide: scope mode based on context
• Ask: major architecture tradeoffs, technology choices
• Gate: if HOLD mode, STOP the pipeline

Phase 3: Engineering Review

Execute /skill:plan-eng-review methodology against the CEO plan:

• Architecture lock-in, data flow, edge cases, test coverage
• Flag: >8 files or >2 new services (complexity check)
• Produce eng review at ~/.gstack/projects/$SLUG/ceo-plans/
• Gate: if architecture issues found, fix before continuing

Phase 4: Design Review (conditional)

If the plan has UI/UX components:

• Execute /skill:plan-design-review methodology
• Score 10 design dimensions 0-10
• Fix the plan for scores < 7

Phase 5: DX Review (conditional)

If the plan has developer-facing surface (API, CLI, SDK, library):

• Execute /skill:plan-devex-review methodology
• Define developer persona, benchmark competitors
• Score 8 DX dimensions, fix scores < 7

Phase 6: Final Gate

Present the complete review summary:

AUTOPLAN COMPLETE
═════════════════
Branch: {branch}
Plan: {plan path}

Pipeline:
  ✅ Office Hours  → {design doc}
  ✅ CEO Review    → {plan} ({mode} mode)
  ✅ Eng Review    → {architecture locked}
  ✅ Design Review → {score}/10
  ✅ DX Review     → {score}/10 ({mode})

Decisions: {N} total ({M} auto-decided, {K} user choices)

Safety Gates: {all clear / issues found}
Recommendation: {PROCEED / PROCEED WITH CAUTION / HOLD}
─────────────────

Decision Audit Trail

Record every decision in the plan file:

## Decision Audit Trail
| # | Phase | Decision | Type | Resolution | Rationale |
|---|-------|----------|------|------------|-----------|
| 1 | CEO | Scope mode | auto | Selective | <20 files, focused feature |
| 2 | CEO | Database choice | user | PostgreSQL | Existing team expertise |
| 3 | Eng | Caching layer | auto | Deferred | Not needed for MVP |

What Gets Auto-Decided

• Scope mode — based on diff size and number of affected components
• Test framework — use what's already in the project
• Trivial architecture choices — when exactly one option fits the constraints
• Deferrals to TODOS.md — non-blocking improvements

What Always Gets Asked

• Technology choices — new language, framework, database, or service
• Architecture tradeoffs — significant coupling/performance/operational decisions
• Design aesthetics — color, typography, layout preferences
• Security-sensitive decisions — auth method, data retention, encryption
• Scope that removes functionality — any HOLD or REDUCTION decision