Sidecar CLI: Simple Workflow Examples

This is a concrete, beginner-friendly workflow for a sample project using Sidecar with an AI coding agent.

The goal is simple: reduce repeated onboarding. Every session starts with context from the last one, and every session leaves useful context for the next one.

Step 1: Initialize Sidecar in a sample project

Example project: todo-api.

 Project setup shell 
 $ mkdir todo-api
$ cd todo-api
$ sidecar init 

Step 2: Prompt your agent to use Sidecar

Copy/paste this as your first prompt after sidecar init:

 Kickoff prompt text 
 Sidecar is initialized in this project. For every task:
1) Run `sidecar context --format markdown` before coding.
2) Implement the requested changes.
3) Record work with `sidecar worklog record --done "<what changed>" --files <paths> --by agent`.
4) If behavior/design changed, record a decision with `sidecar decision record --title "<decision>" --summary "<why>" --by agent`.
5) If follow-up work remains, add a task with `sidecar task add "<follow-up>" --priority medium --by agent`.
6) Run `sidecar summary refresh` at the end.

In your final response, include what you changed and which files were updated. 

Step 3: Open the Sidecar UI (optional, but helpful)

If you want a visual view of tasks, notes, decisions, and recent worklogs:

$ sidecar ui

Keep this open during development so you can quickly scan outstanding work.

Step 4: Craft prompts throughout development

You usually do not need to repeat “use Sidecar workflow” in every prompt if your AGENTS.md or CLAUDE.md already says that. A clear kickoff prompt in Step 2 is enough for most sessions.

Use this simple prompt pattern:

 Task: <what to build>
Constraints: <tests, style, scope>
Definition of done: <what "complete" means> 

Example prompts for the same sample project:

 # Feature prompt
Build a CRUD endpoint for /tasks in todo-api. Include updated files in your final response.

# Bug-fix prompt
Fix the 500 error when creating a task with an empty title. Record a decision if validation behavior changes.

# Refactor prompt
Refactor task validation into a separate module without changing API behavior and list files touched.

# Ask for outstanding tasks
What Sidecar tasks are still outstanding in this project? Summarize them by priority.

# Tell the agent to tackle a specific task
Start with the highest-priority outstanding Sidecar task and complete it end-to-end. In your final response, include what you changed and which files were updated. 

Step 5: Manual commands during development

Even when an agent is doing most coding, you can manually add records at any time:

 Manual updates shell 
 # See current context
$ sidecar context --format markdown

# Add a task yourself
$ sidecar task add "Add pagination to list endpoint" --priority medium --by human

# Add a quick note
$ sidecar note "Need rate limiting before production" --title "API hardening follow-up" --by human

# Record completed work
$ sidecar worklog record --done "Added task filtering by status" --files src/routes/tasks.ts,src/services/tasks.ts --by human

# Record a decision when behavior changes
$ sidecar decision record --title "Reject empty task titles" --summary "Keeps API behavior predictable and avoids invalid data" --by human

# Refresh summary for the next session
$ sidecar summary refresh 

This is the default human + agent workflow most developers use day to day.