Made with :heart: by @fishjojo1
[!NOTE]
RAPID is still in beta. Therefore, features may be broken/unpolished. Feedback is always welcome.
You are deep into a Claude session -- twenty minutes in, the model has seen your architecture,
your test suite, your style conventions. Then it starts repeating itself. It forgets a decision
you made five prompts ago. It hallucinates a function signature it generated correctly earlier.
That is context rot: as the context window fills, the model loses grip on earlier decisions and
artifacts, and quality degrades in ways you cannot predict. RAPID solves this by structuring
work into isolated sets with /clear between every command, so each step gets a fresh context
window loaded with exactly the artifacts it needs -- nothing more, nothing stale.
Install
[!TIP]
Inside Claude Code:
/plugin marketplace add pragnition/pragnition-public-plugins
/plugin install rapid@pragnition-public-plugins
Then,
/rapid:install
Next, restart your terminal/source your shell
Requires Node.js 22+. For alternative installation methods, see DOCS.md.
The /clear Mental Model
After every RAPID command that produces artifacts, a box appears telling you to run /clear and what command comes next. This is intentional -- clearing context between steps keeps each command focused and prevents the degradation that ruins long sessions.
18 of 30 commands show this footer. Informational commands like /rapid:status and /rapid:help do not -- they consume minimal context and produce no artifacts.
Throughout this README, you will see /clear between every command. This is the pattern.
Quickstart
-
/rapid:init -- Research your project, generate a roadmap, decompose work into sets
/clear
-
/rapid:start-set 1 -- Create an isolated worktree for the first set
/clear
-
/rapid:discuss-set 1 -- Capture your implementation vision and design decisions
/clear
-
/rapid:plan-set 1 -- Research, produce wave-level plans, validate contracts
/clear
-
/rapid:execute-set 1 -- Execute all planned waves with parallel agents
/clear
-
/rapid:review 1 -- Scope the set for review (then optionally: unit-test, bug-hunt, uat)
/clear
-
/rapid:merge -- Integrate the completed set into main
/clear
Each command spawns specialized agents, produces artifacts, and advances the set through its lifecycle. The /clear between each step is not optional -- it is what keeps the whole system working.
[!TIP]
RAPID does not confine you to parallel development. Pass --solo to any command to work without worktrees.
Your First Project
This walkthrough assumes you have installed RAPID and have a project you want to work on.
Step 1: Install and configure.
Run /plugin install rapid@pragnition/pragnition-public-plugins , then /rapid:install inside Claude Code. Install configures the RAPID_TOOLS environment variable and builds the agent files your project needs. Reload your terminal/shell to source the new environment variable.
/clear
Step 2: Initialize your project.
Navigate to your project directory in Claude Code and run /rapid:init. RAPID will ask you questions about your project, then spawn 6 research agents in parallel to analyze your codebase. You will be presented with a roadmap of sets -- independent workstreams that can be developed in parallel. Approve or adjust the roadmap before continuing.
/clear
Step 3(optional): Decide on your theming and branding.
Run /rapid:branding. This starts an agent that walks through theming and UI decisions with you, spawning a local server for live testing.
Step 4: Start your first set.
Run /rapid:start-set 1. This creates an isolated copy of your repo (a git worktree) where the first set's work will happen. No changes touch your main branch until you explicitly merge.
/clear
Step 5: Discuss implementation.
Run /rapid:discuss-set 1. RAPID identifies design decisions where multiple valid approaches exist and asks for your input. Your answers are recorded into CONTEXT.md so the planner knows your intent.
/clear
