# Trellis:prep

> Prepare implementation by decomposing a plan into an ordered task breakdown at .specs/{feature}/tasks.json. Use when a plan is ready and you need discrete, verifiable units of work organized in phases.

- Skill: `robmclarty/trellis-prep` (Agent Skill)
- Install (CLI): `npx skillmds@latest add robmclarty/trellis-prep`
- Raw SKILL.md: https://api.skillmd.com/api/skills/robmclarty/trellis-prep/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: robmclarty (https://skillmd.com/u/robmclarty)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/robmclarty/trellis-prep

---


# Prep

Prepare a feature for implementation by decomposing its plan into ordered, verifiable tasks.

## When to use

- "prep this for implementation", "prepare tasks", "break this into tasks"
- "what do I build first", "generate work items", "create task breakdown"
- Any request to decompose a technical plan into actionable steps or prepare work items for execution
- Called automatically by `/trellis:build` when `tasks.json` doesn't exist yet

Create an ordered task breakdown at `.specs/<feature-name>/tasks.json`.

**Recommended effort: medium.** Mechanical decomposition of the plan into ordered tasks.

**Re-running this skill overwrites tasks.json**, resetting all task statuses to "pending". This is how you reset an implementation.

## Pre-flight

Run `python3 ${CLAUDE_PLUGIN_ROOT}/scripts/validate-prereqs.py prep <feature-name>` and use the `specsDir` value from the JSON output. Abort if the output reports missing prerequisites.

## What to ask the user

If the user runs `/prep` without additional context:

1. Are there any ordering constraints beyond what the plan implies? (e.g., "I need the API working before the MCP layer so I can demo it")
2. Should tasks be sized for a human developer or for an AI coding agent? (This affects granularity. Agent tasks can be smaller and more mechanical. Human tasks should be more self-contained.)

Always ask these questions, even when invoked automatically from `/trellis:build` — the user is present and should have the chance to provide input.

## Generation

After gathering all user input, spawn the `task-writer` agent. Pass it: the feature name, specs directory path, and all user-provided context. The agent will read prerequisite files (plan.md, spec.md, guidelines.md) and generate tasks.json.

## Quality gate

After the agent completes, verify the generated `tasks.json`:

- [ ] Valid JSON that conforms to `schemas/tasks.schema.json`
- [ ] Every plan section maps to at least one task
- [ ] Every spec interface has at least one task that implements it
- [ ] Every task has a concrete "verify" field
- [ ] Tasks are ordered so that each task's dependencies are completed in earlier tasks
- [ ] No task requires reading the full plan to understand (it references specific sections)
- [ ] The first phase produces something runnable (even if minimal)
- [ ] The `check` field is populated from guidelines.md (or empty string if not available)