codagent

Codagent Agent Skills

Overview

Codagent is a plugin for Claude Code and Cursor that provides a set of focused skills for spec driven development — from evaluating an idea to implementing with subagents to shepherding a PR through CI. Inspired by obra/superpowers and OpenSpec.

Features

• Evaluate before building — the agent critiques your idea, researches alternatives, and decides if it's worth pursuing
• Interview-driven specs & design — the agent grills you to flesh out requirements, edge cases, and architectural decisions
• Right-sized tasks — breaks work into self-contained task files, each scoped for a single subagent to implement
• Multi-agent implementation — dispatches tasks to be implemented via TDD by Claude Code subagents
• Automated quality gates — Agent Validator runs static checks and AI code reviews for each task before moving on
• End-to-end PR lifecycle — creates the PR, waits for CI, fixes failures, and addresses reviewer comments automatically

Skills

Each skill is invoked with /codagent:<skill-name>.

• init — Initializes Codagent in your project. Checks that the Agent Validator CLI is installed and configured. Safe to re-run.
• propose — Evaluates whether an idea is worth building. Researches the codebase and web, delivers a GO / GO WITH CAVEATS / NO-GO verdict, and writes proposal.md.
• spec — Drives interactive requirement discovery. Walks through each capability from the proposal, asking about behaviors, boundaries, error conditions, and edge cases. Produces spec files with testable WHEN/THEN scenarios.
• design — Creates a technical design through collaborative brainstorming. Proposes 2-3 approaches with trade-offs, presents the design incrementally for approval, and writes design.md.
• review-spec — Reviews design artifacts (proposal, specs, design, tasks) for internal consistency, cross-artifact alignment, and gaps. Reports findings with exact citations.
• plan-tasks — Creates a structured task breakdown from the proposal, design, and specs. Each task file is self-contained with everything a subagent needs to implement it.
• implement-change — Autonomous tech lead that implements a full change end-to-end. Dispatches one implement-and-validate subagent per task (sequentially, never in parallel), runs the gauntlet, archives the change, and finalizes the PR.
• implement-with-tdd — Enforces test-driven development: write a failing test, watch it fail, write minimal code to pass, refactor. No production code without a failing test first.
• implement-and-validate — Autonomous implementer that executes a single task end-to-end. Calls implement-with-tdd to build the code, performs self-review, runs the Agent Validator, and commits on success.
• push-pr — Commits changes, pushes to remote, and creates or updates a pull request. Runs the validator before committing if applicable.
• wait-ci — Polls CI status for the current branch's PR. Enriches failures with GitHub Actions logs, surfaces blocking reviews and unresolved PR comments.
• fix-pr — Fixes CI failures and review comments on the current branch's PR. Dispatches a fixer subagent, verifies with the validator, and pushes.
• finalize-pr — Orchestrates the full post-implementation loop: push PR, wait for CI, fix failures, repeat until green. Stops after 3 fix cycles or when the same failure persists.

Getting Started

1. Install the Agent Validator CLI

Codagent requires the Agent Validator CLI for automated quality verification:

npm install -g agent-validator
agent-validator init

2. Install the plugin

Claude Code

claude plugin marketplace add Codagent-AI/agent-skills
claude plugin install codagent

Cursor

cursor plugins install Codagent-AI/agent-skills

3. Initialize

/codagent:init

Updating

claude plugin marketplace update codagent
claude plugin update codagent@codagent

Then run to get the latest skills:

/codagent:init

License

MIT