CLI Guide

Get the most out of Damper CLI.

A practical guide to project setup, spec writing, project context, completion checklists, and the habits that make AI agents ship reliable code.

Setting Up Your Project

One command to connect Damper to your project.

Project Context

The knowledge base that makes every AI session productive from the first line.

Writing Effective Specs

The anatomy of a task that produces great code.

Quality Gates

Two mechanisms to prevent mistakes and verify quality: critical rules shown at task start, and completion checklists verified at task end.

Setting Up Your Project

One command to connect Damper to your project.

npx @damper/cli setup

This stores your API key in .damper/config.json and adds the Damper MCP server to your Claude Code configuration (~/.claude/settings.json). No code changes required.

First run: Let AI scan your project

After setup, start a Claude Code session and ask it to analyze your codebase. The agent will read your code, identify patterns, and create project context sections that future sessions can reference.

Analyze this codebase and create project context sections
for overview, conventions, testing, and architecture.
Use the Damper MCP tools: update_context_section for each section.

Section-based docs

Organize knowledge into named sections like overview, conventions, testing, or api/architecture. Each section is loaded independently so agents only read what they need.

Hierarchical paths

For monorepos, use paths like api/architecture or api/endpoints. Fetch all children with api/* or all descendants with api/**.

Token-efficient loading

Large sections can be explored block-by-block. Use get_section_blocks to see headings, then get_section_block_content to load only relevant parts.

Managing sections

Use MCP tools to create and update sections. Each section can target specific modules with appliesTo and include tags for discoverability.

Writing Effective Specs

The anatomy of a task that produces great code.

  • Start with an action verb matching the type: "Add ..." for features, "Fix ..." for bugs, "Improve ..." for improvements. A good title tells you what the PR will say.
  • Explain the user problem, the expected behavior, and any constraints. Don't prescribe implementation details here — save that for the plan.
  • Break the work into numbered steps. Reference specific files, functions, or patterns. The more concrete the plan, the less the agent guesses.
  • For features touching multiple modules, create subtasks. Agents check them off as they go, giving you visibility into progress.
  • Tag tasks with labels (backend, frontend, database) and estimate effort (xs, s, m, l, xl). This helps with planning and filtering.
Quality Gates

Two mechanisms to prevent mistakes and verify quality: critical rules shown at task start, and completion checklists verified at task end.

Rules added to context sections via the criticalRules field. They are surfaced automatically when an agent starts a task, so agents can't miss them. Use for patterns that cause real problems when skipped.

update_context_section({
  section: "testing",
  content: "# Test workflow\n\nRun bun test before handoff"
})
Guide

The Complete Workflow

Five steps from zero to shipping code with AI agents.