ainous-team

Ainous Team

A persistent agent team plugin for Claude Code -- 13 roles, 63 skills, that learn and improve over time. v5.20.2.

Built by xdimension.ai

Each role accumulates knowledge across sessions: strategies that work get reinforced, strategies that fail get retired. Trust levels gate permissions so agents earn autonomy through clean performance.

Prerequisites

• Claude Code CLI
• Python 3.6+ (required by hooks)
• bash 3.2+ (macOS stock works)
• git (for knowledge tracking)

Install

# Add the marketplace from GitHub:
claude plugin marketplace add xAlbert1d/Ainous-Team

# Install the plugin:
claude plugin install ainous-team

Updating

# Refresh the marketplace to fetch the latest release:
claude plugin marketplace update xAlbert1d/Ainous-Team

Claude Code picks up the new version on the next session start. If the version doesn't roll over, reinstall:

claude plugin uninstall ainous-team
claude plugin install ainous-team

First-time Setup

After installing the plugin, Claude Code will prompt you to initialize on the next session start. You can also start it manually at any time:

/team-init

This interactive command walks you through choosing your operating mode and then scaffolds ~/.claude/ainous-roles/ with starter files for all 13 roles. The setup is idempotent — running it again won't overwrite existing data.

Mode options:

• Coordinator-as-default (recommended): Claude automatically plans, delegates to role agents, and synthesizes results. You just type tasks naturally.
• Agent mode: You manually invoke roles with @coordinator, @developer, etc.

Alternative — manual setup via shell:

# Coordinator-as-default mode (recommended)
bash "$HOME/.claude/plugins/marketplaces/ainous-team/scripts/setup.sh"

# Agent mode
bash "$HOME/.claude/plugins/marketplaces/ainous-team/scripts/setup.sh" --agentmode

Post-commit journal reminder for coordinators: run bash scripts/install-post-commit-journal-reminder.sh once per repo to enable automatic commit-time prompts.

Before any release, run bash scripts/verify-role-infrastructure.sh to confirm all roles have complete 4-file scaffold (playbook, growth, journal, learnings) + agent definition + capability card.

Coordinator-as-Default Mode

By default, setup configures Claude Code to be the coordinator -- it will automatically plan, delegate to role agents, and synthesize results. You just give it tasks naturally:

# Claude automatically plans, spawns roles, and synthesizes:
implement user auth for the API    # spawns researcher + architect + developer + tester
fix the login bug                  # spawns developer + tester
scan src/ for vulnerabilities      # spawns security

To disable later, remove the "You ARE the Coordinator" section from ~/.claude/CLAUDE.md.

Agent Mode (--agentmode)

If you prefer to invoke roles explicitly:

@coordinator implement user auth for the API    # full team orchestration
@developer fix the login bug                     # direct role invocation
@security scan src/ for vulnerabilities          # security audit
/team-status                                     # dashboard
/team-history                                    # session history
/team-alerts                                     # health checks
/team-retro                                      # periodic team review

How It Works

User --> @coordinator --> plans task --> spawns role teammates --> synthesizes results
                           |                  |                        |
                      @authority         contracts +              verify gates
                   (approval gate)     playbook injection      (loop on failure)

1. User invokes @coordinator with a task
2. Coordinator selects topology -- fast-fix for bugfixes, security-first for auth, full-pipeline for features, etc. Then assesses parallelizability and chooses spawn mode (Agent for quick tasks, tmux for long-running).
3. Coordinator spawns teammates -- each gets their playbook + project context + team knowledge + execution contract. Can spawn multiple instances of the same role for independent tasks.
4. Verification gates between phases -- tests must pass, findings must be actionable, contracts must be fulfilled
5. Teammates work -- message @authority for approvals, @security for scans
6. Coordinator synthesizes -- presents unified result, asks for rating
7. On stop -- coordinator runs team retrospective (cross-role interaction analysis) and detects user corrections (implicit feedback from manual edits)

The coordinator has Write/Bash for journal writes but delegates all implementation to role agents. It has no Edit tool. It must delegate to @developer for code, @writer for docs, @tester for tests.

Architecture