agent-team

AI Coding Configurations

An opinionated collection of Cursor and Claude Code configurations for AI-assisted development, with project-local pi maintainer skills for editing the workflows safely. Drop the relevant folder into any project to get structured workflows, specialized subagents, safety hooks, and reusable skills - all working together out of the box.

---

Install

Two paths, pick one:

Option A — Plugin (recommended for sharing)

/plugin marketplace add mike-diff/ai-coding-configs
/plugin install agent-team@mike-diff

Commands surface as /agent-team:discuss, /agent-team:dev, etc. See plugin README for details.

Option B — Standalone (drop-in, unprefixed commands)

git clone https://github.com/mike-diff/ai-coding-configs.git
cp -r ai-coding-configs/.claude/ /path/to/your/project/

Commands surface as /discuss, /dev, etc. See .claude/README.md for details.

Working on this repo with pi

This repository includes project-local pi maintainer skills:

/skill:agent-team-discuss
/skill:agent-team-spec
/skill:agent-team-dev

These are not another product surface. They are operator workflows for safely editing and validating the Agent Team configs across .claude/, plugins/agent-team/ and .cursor/.

Note: pi v0.79.0+ prompts for project trust before loading .pi/skills/. Approve the prompt (or run with --approve, or manage saved decisions via /trust) for these skills to appear.

---

What's Included

Tool	Config Folder
Claude Code	.claude/
Cursor	.cursor/

Both share the same core philosophy and command set, with each adapted to their platform's native capabilities.

---

How It Works

Each configuration gives your AI assistant a set of commands to run, subagents to delegate work to, rules to follow, hooks to enforce automatically, and skills to draw on as reference.

Commands

Commands are slash commands you run to kick off a workflow. Both tools share the same set, implemented as Agent Skills following the agentskills.io specification:

Command	What it does
/discuss	Think through an idea before building. Spawns background research, interviews you, validates the plan, and emits an ADLC handoff.
/spec	Turn a feature description into a spec-backed contract: requirement validation, architecture planning, architecture validation, and phased tasks.
/dev	Build a feature end-to-end with a coordinated subagent team, then reflect, review/QA, commit or report PR-ready, and wrap up learnings.
/to-dos	Break down a feature into detailed, dependency-tracked tasks.
/issue	Fetch a GitHub issue, explore the codebase, produce an implementation plan.
/ticket	Create a well-structured GitHub issue through a guided interview.
/orient	Map the tech stack, architecture, and patterns of an unfamiliar codebase.
/ask	Ask clarifying questions before proceeding with work.
/skill	Create a new skill using TDD - baseline test, write content, validate against the agentskills.io spec.
/slop-check	Run tool-driven code quality analysis and conservative cleanup judgment.
/primitives	List every native tool and capability available in the current session.

---

Subagents

Subagents are specialized agents that the orchestrator delegates work to. Each has a narrow focus.

Claude Code (.claude/agents/):

Agent	Role
explorer	Read-only codebase analysis
implementer	Writes and modifies code
reviewer	Spec compliance and code quality
qa	Runs lint, typecheck, and tests
skill-author	Creates skills using TDD

Cursor (.cursor/agents/):

Agent	Role
explorer	Codebase analysis
implementer	Code implementation
spec-reviewer	Spec compliance verification
checker	Lint and typecheck
tester	Test execution
browser-tester	UI verification
skill-author	Skill creation via TDD

---

Rules

Rules are guidelines loaded automatically by the AI at the start of every session. They can't be ignored the way inline instructions sometimes are.

Claude Code (.claude/rules/):

Rule	What it covers
coding-standards	Code quality, naming, structure
mcp-caching	Cache large MCP responses to .context/ to avoid bloating context windows

Cursor (.cursor/rules/):

Rule	What it covers
coding-standards	Code quality, naming, structure
dev-workflow	Orchestration conventions for /dev
commit-conventions	Conventional Commits format
subagent-outputs	Required result block formats for subagents
mcp-caching	Cache large MCP responses to .context/

---