diegoboni/aiagents

    _    ___    _                    _
   / \  |_ _|  / \   __ _  ___ _ __ | |_ ___
  / _ \  | |  / _ \ / _` |/ _ \ '_ \| __/ __|
 / ___ \ | | / ___ \ (_| |  __/ | | | |_\__ \
/_/   \_\___/_/   \_\__, |\___|_| |_|\__|___/
                    |___/

A reusable AI agent framework designed to be bootstrapped into any software project. Provides a structured spec-driven pipeline with multi-agent support, harness engineering, and domain-scoped skills for Claude, Codex, and Gemini — keeping token usage minimal by loading only the context relevant to each task.

---

Install

Run this from inside any project you want to set up:

bash <(curl -fsSL https://raw.githubusercontent.com/DiegoBoni/AIAgents/main/install.sh)

With options:

# Only Claude
bash <(curl -fsSL https://raw.githubusercontent.com/DiegoBoni/AIAgents/main/install.sh) --agent claude

# Specific target folder
bash <(curl -fsSL https://raw.githubusercontent.com/DiegoBoni/AIAgents/main/install.sh) --target /path/to/project

# All agents, symlink mode (updates automatically when this repo changes)
bash <(curl -fsSL https://raw.githubusercontent.com/DiegoBoni/AIAgents/main/install.sh) --agent all --mode link

After install, open the project in your agent and run /scan to get started.

---

The problem this solves

Most AI agent setups load the entire project context on every call. As a project grows, this becomes expensive and noisy. This kit solves that with domain skills: each skill loads only the section of the project context it needs.

Without this kit:        With this kit:
─────────────────────    ──────────────────────────────
Every task loads:        Backend task loads:
  - Full stack info        - [context.backend] only
  - Frontend config        ~300 words, not 3000
  - DB schemas
  - CI/CD pipelines
  - All standards
  (~3000 words)

---

How it works

1. One context file, five domain sections

After running /scan, the file .ai/project-context.md contains a global summary and five independently loadable domain sections:

.ai/project-context.md
├── Metadata / Stack / Architecture   (global — loaded by /scan and /spec)
├── [context.backend]    ← loaded by backend skill only
├── [context.frontend]   ← loaded by frontend skill only
├── [context.data]       ← loaded by data skill only
├── [context.testing]    ← loaded by testing skill only
└── [context.devops]     ← loaded by devops skill only

2. Active spec pointer — .ai/current

Every spec-related command reads .ai/current automatically. No need to pass file paths between commands.

.ai/
├── project-context.md     ← populated by /scan
├── current                ← "specs/features/checkout-flow"  (set by /spec)
├── log/
│   └── session.log        ← audit trail (populated by /harness hooks)
└── scripts/
    └── on-session-end.sh  ← runs after every Claude Code session

specs/
├── features/
│   ├── checkout-flow/