glossary
Key terms and concepts you'll encounter when using twig.
agent
The AI assistant powered by Claude that can autonomously read code, make edits, and run commands. Unlike traditional code completion tools that only suggest changes, the agent can take actions on your behalf after you approve them.
workspace
An isolated development environment based on Git worktrees. Each workspace lets you work on a separate task without conflicts. Twig creates and manages workspaces automatically, allowing you to switch between different features or bug fixes without affecting each other.
context
Information the agent uses to understand your request and generate appropriate responses. Context includes:
- The conversation history
- Files you've opened or referenced
- Code the agent has read
- Your project structure
The agent automatically manages context, but you can help by being specific about what's relevant.
posthog integration
Twig's connection to your PostHog analytics account. This integration allows the agent to access product data, helping you make development decisions based on real user behavior rather than assumptions.
approval flow
Twig's review process for changes. Before any code is modified, the agent shows you:
- What file will change
- A diff of the proposed changes
- Why the change is being made
You can approve, reject, modify, or ask questions before changes are applied.
terminal
The built-in command-line interface at the bottom of twig. Use it to run tests, start dev servers, execute Git commands, or any other terminal operations without leaving the editor.
rules
Persistent instructions that guide the agent's behavior across conversations. Rules let you encode your project's standards, coding patterns, and preferences so you don't have to repeat them in every conversation.
Rules can be project-specific (shared with your team) or personal (just for you).
model context protocol (mcp)
A standard for connecting AI agents to external tools and data sources. Through MCP, twig's agent can access databases, APIs, documentation systems, and other resources beyond your local codebase.
MCP requires configuration in Claude, not in twig's interface.
prompting
How you communicate your intent to the agent. Effective prompting is specific, provides context, and clearly states what you want to accomplish. Vague requests lead to vague results.
conversation
An ongoing dialogue with the agent. Conversations maintain context and history, but you can start fresh conversations when switching to unrelated topics.
related pages
- agentic development - understand the agent's capabilities
- working with the agent - learn to prompt effectively
- rules - set up persistent instructions