# Quickstart Demo This demo shows what dev-flow adds to a normal coding-agent session. ## Starting Point You have a project and want an agent to implement a feature without losing the requirement, task list, test status, or iteration notes. Without dev-flow, the prompt usually has to carry everything: ```text Add a settings page. Think through requirements, make a plan, implement it, test it, and remember what changed. ``` That works for small changes, but it becomes fragile when the task spans many files or multiple sessions. ## Install ```bash curl -fsSL https://raw.githubusercontent.com/daphnee-ovo/dev-flow/main/install/install.sh | bash # follow the interactive dev-flow setup cd your-project ``` ## Run The Workflow Ask the agent: ```text /init ``` dev-flow creates the current branch's workflow workspace: ```text .dev-doc/ └── / ├── STATUS.yaml ├── CHANGELOG.md ├── task/ └── issue/ VERSION ``` Phase documents (BRAINSTORM.md, PRD.md, SPEC.md) are created by their respective commands as needed. Then ask for a task breakdown: ```text /task ``` The agent writes structured task files under `.dev-doc//task/` instead of keeping the plan only in chat. During development, dev-flow hooks keep the agent aligned with the workflow: - context injection reminds the agent of current phase and document rules - guard hooks block unsafe writes such as system temp paths - post-write hooks detect doc/code drift - stop hooks preserve changelog context ## What You Get After one feature loop, the repository contains both code changes and a lightweight delivery record: ```text .dev-doc/ ├── archive.db # SQLite history queried by dow archive ... └── main/ ├── STATUS.yaml # current phase, mode, version, active task ├── CHANGELOG.md # session-level change notes ├── task/ # planned work before iterate └── issue/ # discovered defects before iterate ``` The point is not ceremony. The point is to make long-running agent work inspectable, recoverable, and harder to fake. For concrete sample output, inspect [sample-project](sample-project/).