snap-plan
Snap Plan
Read a GitHub issue → identify durable decisions → draft tracer-bullet phases → post as issue comment. Issue-type agnostic — works on PRDs or individual slices; plan depth scales to issue scope.
1. Get the issue
If issue URL/number in conversation context, use it. Otherwise ask. Fetch
with gh issue view <number>.
2. Explore codebase
Understand current architecture, existing patterns, integration layers. Focus on what's relevant to the issue — module boundaries, dependency graphs, test infrastructure.
Use subagents for deep exploration of unfamiliar areas.
3. Identify durable architectural decisions
Before phasing, lock down decisions unlikely to change across implementation:
- Route structures / URL patterns
- Database schema shape
- Key data models and abstractions
- Auth/authz approach
- Third-party service boundaries
These anchor the plan — every phase references them.
4. Draft tracer-bullet phases
Break the issue into vertical slices. Each phase is a thin end-to-end path through ALL layers — never a horizontal slice of one layer.
- Each slice delivers a narrow but COMPLETE path (schema → API → logic → UI → tests)
- Completed slice is demoable or verifiable standalone
- Prefer many thin slices over few thick ones
- Include durable decisions (routes, schemas, model names) — exclude volatile details (file paths, function names)
- Order simplest-to-most-complex; first phase is the tracer bullet (thinnest possible end-to-end proof)
5. Quiz user
Present breakdown as numbered list. Per phase: title, what it covers, what it builds, acceptance criteria.
Ask: granularity right? Merge or split any phases? Iterate until approved.
6. Post as issue comment
Post approved plan as comment on the source issue via gh issue comment.
## Durable Decisions
- **Routes:** ...
- **Schema:** ...
- **Key models:** ...
---
## Phase 1: [Title] (tracer bullet)
**Covers:** [user stories / requirements from source issue]
### What to build
[End-to-end behavior, not layer-by-layer]
### Acceptance criteria
- [ ] ...
---
## Phase 2: [Title]
...
Principles
- Vertical over horizontal. Every phase is end-to-end, never one-layer-deep.
- Durable over volatile. Plan references routes/schemas/models, not file paths.
- Tracer bullet first. Thinnest possible proof that the path works.