Planning Overview
How Trovella plans work, makes decisions, runs retrospectives, and coordinates AI agent development.
Trovella is built by a solo founder (Kyle) working part-time alongside a full-time job. Every planning process is designed around three constraints: limited human hours, high AI agent leverage, and the need for decisions to persist across stateless agent sessions.
This topic covers the methodology behind planning -- not the specific technical decisions themselves, which live in each technical domain's ADR pages.
Core Principles
Product decisions happen just-in-time. They are made in the evening before the technical work they inform, so agents can execute the next morning. Speculative decisions that sit unused create stale context.
Progressive complexity. Start simple, upgrade at explicit triggers. Infrastructure, tools, and processes are added when a concrete need exists -- not when a hypothetical future need is imagined.
Documentation is the memory layer. Every Claude Code session starts fresh. Decisions, patterns, and conventions survive only through written artifacts: CLAUDE.md files, ADRs, guides, and retrospectives. Planning outputs must be written down or they are lost.
Milestones over sprints. Fixed weekly sprint commitments create guilt on variable-schedule weeks. Milestones (Linear Projects) focus energy on outcomes. Weekly cycles are data-collection check-ins for capacity learning, not commitments.
Planning Artifacts
| Artifact | Purpose | Location |
|---|---|---|
| Phase plan | Goals, scope, and sequencing for a multi-week phase | docs/ (per-phase documents) |
| Week 0 Decisions | Initial technology and architecture decisions | docs/Week 0 - Initial Decisions.md |
| ADRs | Individual architecture decision records | docs/architecture/decisions/ |
| Implementation retrospective | What worked, what broke, proposed changes | docs/phase-0/Phase 0 - Implementation Retrospective.md |
| Linear milestones | Target outcomes with child issues and dates | Linear (source of truth for task tracking) |
| Weekly cycle | Issues pulled from milestone backlogs for the week | Linear Cycles (Mon--Sun) |
| CLAUDE.md hierarchy | Agent-facing rules and conventions | Root + per-package CLAUDE.md files |
| Guides | Detailed how-to references for specific domains | docs/*.md |
Pages in This Topic
- Phase Approach -- how phases are defined, scoped, and sequenced
- Decision-Making Process -- the Week 0 decision sprint, ADR format, and ongoing decision patterns
- Retrospective Format -- how implementation retrospectives work and what they produce
- Agent Workflow (SPTI) -- the Specify-Plan-Tasks-Implement workflow for AI agent development
- Sprint Cadence & Sustainability -- weekly planning, milestones, throughput tracking, and burnout prevention
Cross-Domain References
- Product Vision -- what we are building and for whom
- Conventions -- process conventions (commit rules, ticket format, code review)
- Delivery -- Pipeline -- CI/CD pipeline that enforces quality gates
- Delivery -- Architecture Enforcement -- automated checks that make architectural decisions stick