squads-cli 0.2.0 → 0.2.1

package/README.md

@@ -1,407 +1,640 @@
-# squads-cli
+<div align="center">
 
-[![npm version](https://img.shields.io/npm/v/squads-cli)](https://www.npmjs.com/package/squads-cli)
+# Agents Squads
+
+**Your AI ops teams.**
+
+Autonomous AI agents for engineering, marketing, finance, and operations.
+You make the decisions. They do the work.
+
+[![npm version](https://img.shields.io/npm/v/squads-cli.svg)](https://www.npmjs.com/package/squads-cli)
+[![npm downloads](https://img.shields.io/npm/dw/squads-cli.svg)](https://www.npmjs.com/package/squads-cli)
 [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE)
-[![Node.js](https://img.shields.io/badge/node-%3E%3D18-brightgreen)](https://nodejs.org)
+[![Node.js](https://img.shields.io/badge/node-%3E%3D18-brightgreen.svg)](https://nodejs.org)
+[![GitHub stars](https://img.shields.io/github/stars/agents-squads/squads-cli?style=social)](https://github.com/agents-squads/squads-cli)
 
-**Organize, run, and track autonomous AI agents.** Built for Claude Code.
+</div>
 
-```
-$ squads status
-
-  squads status
-  ● 7 active sessions across 1 squad (claude 7)
-
-  10/10 squads  │  memory: enabled
-
-  ┌────────────────────────────────────────────────────────┐
-  │ SQUAD           AGENTS  MEMORY        ACTIVITY         │
-  ├────────────────────────────────────────────────────────┤
-  │ cli             7       1 entry       today            │
-  │ engineering     6       1 entry       today            │
-  │ intelligence    17      1 entry       4d ago           │
-  │ marketing       4       2 entries     today            │
-  │ website         10      1 entry       5d ago           │
-  └────────────────────────────────────────────────────────┘
-```
+## Why Squads?
 
-## Why squads-cli?
+Agents Squads is an experimental framework that replicates how businesses
+have been organized and built over decades of evolution — different
+domains, specialized teams, shared goals, prioritization, pivoting, and
+accumulated knowledge. The same structure that makes human organizations
+effective can make AI operations effective.
 
-AI agents are powerful individually. But real work requires coordination.
+LLMs are not learners. Their limitations around self-learning and memory
+are well known. But this framework engineers around those limitations —
+using structured context injection, persistent filesystem memory, and
+feedback loops to extract the best outputs from repetitive tasks and
+accumulate organizational knowledge across cycles.
 
-- **Squads** — Group agents by domain (engineering, research, marketing)
-- **Memory** — Persistent state that survives across sessions
-- **Goals** — Track objectives and measure progress
-- **Sessions** — Real-time detection of running AI assistants
-- **Stack** — Local infrastructure for telemetry and memory
+AI agents from different providers — Claude, Gemini, Codex, Grok — each
+have distinct strengths. Claude reasons deeply. Gemini is fast and cheap.
+GPT has broad knowledge. But individually, each one hits a ceiling: no
+shared context, no coordination, no way to evaluate its own output.
 
-No complex infrastructure. Just markdown files and a CLI.
+A **squad** puts multiple AI agents together as a team — different models,
+different roles — working toward a shared goal. A Gemini scanner finds
+opportunities. A Claude worker executes deep reasoning. A fast model
+verifies quality. A lead coordinates and prioritizes. The synergy between
+models and roles produces output no individual agent achieves alone.
 
-## Installation
+Agents within a squad don't just run in parallel — they **converse**.
+A lead briefs the team, workers iterate on the task, the lead reviews
+and redirects, the verifier checks the output. This local conversation
+loop — happening entirely on your machine through shared transcript
+files — lets agents discuss, debate, and converge on a solution before
+shipping anything.
 
-```bash
-npm install -g squads-cli
-```
+This is the difference between "AI that helps" and "AI that operates."
+
+### Who uses this CLI
+
+`squads run` is called by an orchestrating agent — Claude Code, Gemini,
+or any supported CLI. After dispatch, **agents are the primary users of
+this CLI**. They call `squads memory read` to recall what they know,
+`squads env show --json` to understand their execution context, and
+`squads status --json` to see what's happening across the org.
+
+The CLI is designed for both human operators and machine consumers —
+every command supports `--json` for programmatic access. Human-facing
+commands (like `squads dash`) prioritize readability. Agent-facing
+commands (like `squads env prompt`) prioritize composability.
+
+### Context is the moat
+
+Most agent frameworks focus on tool calling. Squads focuses on **what the
+agent knows before it starts working**.
+
+Every agent execution loads a layered context cascade — squad identity,
+current priorities, feedback from last cycle, active work across the team —
+tuned by role so scanners stay lightweight and leads get the full picture.
+
+The desired result: agents that don't duplicate work, don't ignore
+feedback, and improve with every cycle.
 
 ## Quick Start
 
 ```bash
-# Initialize in your project
+npm install -g squads-cli
 squads init
-
-# See what you have
 squads status
+```
 
-# Full dashboard with goals and metrics
-squads dash
+`squads init` creates a `.agents/` directory with 4 starter squads and
+configures your environment.
+
+## How It Works
+
+Everything in Squads is a file. There's no database, no server, no
+runtime to manage. Your entire AI workforce lives in a single `.agents/`
+directory that you commit to git like any other code. This means you get
+version history, code review, branching, and merge — applied to your
+agent definitions and organizational memory.
+
+```
+.agents/
+├── BUSINESS_BRIEF.md          # Business context (primary source)
+├── config/
+│   └── SYSTEM.md              # Base behavior (shared across all agents)
+├── squads/
+│   ├── intelligence/
+│   │   ├── SQUAD.md            # Squad identity, goals, KPIs
+│   │   ├── scanner.md          # Agent definition
+│   │   └── analyst.md          # Agent definition
+│   ├── research/
+│   │   ├── SQUAD.md
+│   │   └── analyst.md
+│   ├── product/
+│   │   ├── SQUAD.md
+│   │   └── scanner.md
+│   └── company/
+│       ├── SQUAD.md
+│       └── evaluator.md        # COO — evaluates all squad outputs
+└── memory/                     # Persistent state (auto-managed)
+    ├── intelligence/
+    ├── research/
+    ├── product/
+    └── company/
+```
+
+A squad is a directory. An agent is a markdown file. Edit in vim, review in
+a PR, diff in git. No YAML pipelines, no JSON schemas, no DSLs.
+
+The three top-level directories serve distinct purposes: `config/` holds
+immutable rules that every agent follows regardless of squad. `squads/`
+holds identity — who each agent is, what it produces, how it behaves.
+`memory/` holds state — what agents have learned, what they're working on,
+what feedback they've received. Identity is stable. Memory evolves.
+
+### Context Cascade
+
+The biggest problem with autonomous agents isn't capability — it's
+context. An agent that doesn't know what's already been done will
+duplicate work. An agent that doesn't know the company strategy will
+optimize for the wrong thing. An agent that doesn't know its own last
+output was rated as noise will keep producing noise.
+
+Squads solves this by loading a layered context cascade before every
+execution, in strict priority order:
+
+| # | Layer | Source | Purpose |
+|---|-------|--------|---------|
+| 0 | System Protocol | `config/SYSTEM.md` | Immutable rules every agent follows |
+| 1 | Squad Identity | `squads/{squad}/SQUAD.md` | Mission, goals, output format |
+| 2 | Priorities | `memory/{squad}/priorities.md` | Current operational focus |
+| 3 | Directives | `memory/company/directives.md` | Company-wide strategic overlay |
+| 4 | Active Work | `memory/{squad}/active-work.md` | Open PRs and issues — prevent duplication |
+| 5 | Agent State | `memory/{squad}/{agent}/state.md` | What the agent already knows |
+| 6 | Feedback | `memory/{squad}/feedback.md` | Last cycle evaluation |
+| 7 | Briefings | `memory/daily-briefing.md` | Cross-squad context |
+
+The order matters. When the token budget runs out, lower layers drop
+first. An agent that loses briefings still knows its mission and what
+work exists. An agent that loses its identity is useless. The cascade
+ensures graceful degradation — the most critical context always loads.
+
+### Role-Based Depth
+
+Not every agent needs the same depth of context. A scanner looking for
+new opportunities doesn't need to know the company's strategic directives
+or what feedback other agents received. A lead coordinating across squads
+needs everything. Loading unnecessary context wastes tokens and can
+confuse the agent with irrelevant information.
+
+- **Scanners** get identity, priorities, and their own state — they discover, don't decide
+- **Workers** add directives, feedback, and active work — they execute with awareness of what exists
+- **Leads** get all layers including cross-squad briefings — they orchestrate with full visibility
+- **Evaluators** get all layers with org-wide summaries — they assess and generate feedback
+
+### Goals vs Priorities
+
+A common failure mode in autonomous systems is conflating direction with
+execution. An agent that only sees "Fix #461 this week" doesn't know
+*why* that matters or what the bigger picture is. An agent that only sees
+"Build the best developer experience" has no idea what to work on today.
+
+Squads separates aspiration from execution:
+
+- **Goals** live in `SQUAD.md` — atemporal, aspirational ("Zero friction first-run experience")
+- **Priorities** live in `priorities.md` — temporal, operational ("Fix #461 this week")
+
+Goals give agents purpose and judgment. Priorities give them focus. Both
+are injected — goals as identity context that shapes decision-making,
+priorities as immediate operational focus. `squads goal set` writes
+aspirational goals. Priorities are updated between cycles by the human
+operator or the evaluator agent.
+
+### Phase Ordering
+
+In a real organization, the research team finishes their analysis before
+the product team writes the roadmap. The finance team closes the books
+before the CEO reviews performance. Order matters — and when agents
+execute in the wrong order, they work with stale or missing information.
+
+Squads declare dependencies in their SQUAD.md frontmatter:
+
+```yaml
+---
+name: product
+depends_on: [intelligence, research]
+---
+```
 
-# Run a squad
-squads run engineering
+The CLI computes execution phases via topological sort. Squads with no
+dependencies run first. Squads with `depends_on: ["*"]` run last
+(evaluation). Within each phase, squads run in parallel.
 
-# Search memory
-squads memory query "authentication"
+### The Feedback Loop
 
-# Set a goal
-squads goal set engineering "Ship v2.0 by Friday"
+This is the core of Squads — a closed loop where agents improve autonomously:
+
+```
+┌─────────────────────────────────────────────────────┐
+│                                                     │
+│   autopilot ──→ squads run                          │
+│       ▲              │                              │
+│       │              ▼                              │
+│       │    intelligence ──┐                         │
+│       │    research ──────┼──→ product              │
+│       │                   │        │                │
+│       │                   │        ▼                │
+│       │                   │   company (COO)         │
+│       │                   │        │                │
+│       │                   │   feedback.md           │
+│       │                   │        │                │
+│       │                   └────────┘                │
+│       │               (injected next cycle)         │
+│       └─────────────────────┘                       │
+│                                                     │
+└─────────────────────────────────────────────────────┘
 ```
 
-## Features
+After each cycle, the company evaluator assesses all squad outputs:
+what was valuable, what was noise, what to prioritize next. Written to
+`feedback.md` per squad and injected into the next cycle — closing the
+loop so agents learn from their own output quality.
 
-### Dashboard
+`squads autopilot` uses these evaluations to determine which squads to
+run next, in what order, with what budget. The full loop: autopilot
+dispatches → agents execute → evaluator writes feedback → autopilot
+reads feedback → dispatches again.
 
-```
-$ squads dash
+## Running Agents
 
-  squads dashboard
-  ● 7 active sessions across 1 squad (claude 7)
+There are three ways to run agents, each suited for different situations.
 
-  8/10 squads  │  404 commits  │  use -f for PRs/issues
+**Single agent** — run one agent for a focused task. The agent gets its
+context cascade, executes autonomously, and writes results to GitHub
+and memory. This is the building block.
 
-  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 35% goal progress
+```bash
+squads run research/analyst
+squads run intelligence --task "Scan competitor pricing changes"
+```
 
-  ┌──────────────────────────────────────────────────────────┐
-  │ SQUAD        COMMITS PRs ISSUES GOALS  PROGRESS          │
-  ├──────────────────────────────────────────────────────────┤
-  │ marketing    203     0   0/0    9/12   ━━━━━━━━          │
-  │ website      203     0   0/0    0/1    ━━━━━━━━          │
-  │ engineering  139     0   0/0    0/1    ━━━━━━━━          │
-  │ cli          48      0   0/0    2/3    ━━━━━━━━          │
-  └──────────────────────────────────────────────────────────┘
+**Squad conversation** — run an entire squad as a coordinated team. The
+lead briefs first, workers execute in parallel, the lead reviews outputs,
+and the cycle repeats until the team converges on a result. This is where
+multi-agent synergy happens.
 
-  Git Activity (30d)
-  Last 14d: ▁▁▁▁▁▁▁▄▆▄▆▅█▂
-  404 commits  │  13.5/day  │  21 active days
+```bash
+squads run research --parallel
 ```
 
-### Memory Search
+**Autonomous dispatch** — let Squads decide what to run, when, and in
+what order. Autopilot reads priorities and feedback, respects phase
+ordering, and manages budget constraints. This is the hands-off mode for
+continuous operations.
 
+```bash
+squads autopilot --interval 30 --budget 50
 ```
-$ squads memory query "telemetry"
 
-  squads memory query "telemetry"
+## Starter Squads
+
+The first experience matters. Rather than shipping an empty framework and
+asking you to figure out what to build, `squads init` ships with 4 squads
+designed to deliver visible output from the very first run. These aren't
+demo squads — they're the foundation of an operational AI workforce.
+
+| Squad | What It Does | Agents |
+|-------|-------------|--------|
+| **intelligence** | Strategic synthesis — Know / Don't Know / Playbook briefs | intel-lead, intel-eval, intel-critic |
+| **research** | Market, competitor, and trend research with sourced findings | lead, analyst, synthesizer |
+| **product** | Roadmap, specs, user feedback synthesis | lead, scanner, worker |
+| **company** | Orchestrates squads, evaluates outputs, closes the feedback loop | manager, evaluator, goal-tracker, event-dispatcher, critic |
 
-  5 results found
+**Intelligence + research** produce insights. **Product** turns insights
+into roadmap. **Company** evaluates everything and writes feedback — which
+agents read next cycle. The loop closes automatically.
 
-  ┌──────────────────────────────────────────────────┐
-  │ LOCATION                    TYPE      SCORE     │
-  ├──────────────────────────────────────────────────┤
-  │ cli/cli-lead                state     7.2       │
-  │ engineering/eng-lead        state     7.2       │
-  │ marketing/marketing-lead    state     7.2       │
-  └──────────────────────────────────────────────────┘
+Additional squads available as packs:
 
-  Matches
-  ◇ Telemetry pipeline COMPLETE. Dashboard showing real-time...
-    └ cli/cli-lead
+```bash
+squads init --pack engineering    # Add engineering squad
+squads init --pack marketing      # Add marketing squad
+squads init --pack all            # All available squads
 ```
 
-### Session Detection
+### Build Your Own
 
-Real-time detection of running AI coding assistants:
+Squads are directories. Agents are markdown files. Bring your own skills,
+tools, and CLIs — anything your agents can run in a terminal becomes part
+of the squad.
 
+```bash
+squads add devops -d "Infrastructure and deployment automation"
 ```
-$ squads status
 
-  ● 7 active sessions across 1 squad (claude 7)
+This creates `squads/devops/SQUAD.md` and a lead agent. Add more agents
+as `.md` files, define their roles, give them access to your CLIs
+(`terraform`, `kubectl`, `aws`, `docker` — whatever they need), and run:
+
+```bash
+squads run devops --parallel
 ```
 
-Supports multiple tools:
-- Claude Code
-- Cursor
-- Aider
-- Gemini
-- GitHub Copilot
-- Sourcegraph Cody
-- Continue
+### Customizing Your Squads
+
+Agents read context in layers — higher layers are stable, lower layers
+change often. Update them in this order:
 
-### Stack Management
+#### 1. Business Brief (what you do)
 
-Local Docker infrastructure for telemetry and memory:
+Edit `.agents/BUSINESS_BRIEF.md` — every agent reads this before every run.
+The more specific you are, the better agents perform.
 
+```markdown
+What does your business do? Who are your customers? What market?
+What should agents research first? Who are your competitors?
 ```
-$ squads stack health
 
-  squads stack health
+#### 2. Directives (what matters now)
 
-  ✓ postgres   healthy
-  ✓ redis      healthy
-  ✓ neo4j      healthy
-  ✓ bridge     healthy
-  ✓ langfuse   healthy
-  ✓ mem0       healthy
-  ✓ engram     healthy
+Edit `.agents/memory/company/directives.md` — strategic overlay that
+overrides squad-level goals when there's a conflict.
 
-  ● 8/8 services healthy
+```markdown
+What is the #1 priority right now? What metric are you optimizing?
+What constraints apply? What should agents NOT do?
 ```
 
-### Auto-Update
-
-```
-$ squads status
+#### 3. Squad Goals (what each team does)
 
-  ⬆ Update available: 0.1.2 → 0.2.0 (run `squads update`)
+Replace the generic goals in each `SQUAD.md` with goals specific to your
+business:
 
-$ squads update
-  Checking npm registry...
-  ⬆ Update available: 0.1.2 → 0.2.0
-  Update now? [y/N]: y
-  Installing update...
-  ● Updated to 0.2.0
+```bash
+# Or use the CLI:
+squads goal set intelligence "Monitor competitor X's pricing weekly"
+squads goal set research "Deep dive on Y market segment"
+squads goal set product "Write spec for Z feature"
 ```
 
-## Core Concepts
+#### 4. Priorities (what to do this week)
 
-### Squads = Domain-Aligned Teams
+Create or edit `.agents/memory/{squad}/priorities.md` — operational focus
+that changes frequently:
 
-```
-.agents/squads/
-├── engineering/
-│   ├── SQUAD.md           # Squad config + goals
-│   └── ci-optimizer.md    # Agent definition
-├── research/
-│   ├── SQUAD.md
-│   └── market-analyst.md
-└── intelligence/
-    └── ...
+```markdown
+- Fix issue #123 (blocking users)
+- Research competitor's new feature launch
+- Update roadmap based on last cycle's feedback
 ```
 
-### Agents = Markdown Prompts
+**Rule**: Goals are aspirational (stable). Priorities are operational
+(updated frequently). Directives are strategic (updated less frequently).
 
-```markdown
-# CI Optimizer
+#### Agent Instructions
 
-## Purpose
-Reduce build times and optimize CI/CD pipelines.
+Each `.md` file in a squad defines an agent's behavior, output format,
+and quality rules. The starter agents come with structured output formats
+(tables, scoring rubrics, required sections) — modify these to match what
+you actually need.
 
-## Model
-claude-sonnet-4
+#### System Protocol
 
-## Tools
-- Bash(gh:*, git:*)
-- Read
-- Edit
+`.agents/config/SYSTEM.md` contains immutable rules all agents follow —
+git workflow, memory protocol, output standards. You rarely need to change
+this, but you can customize it for your team's conventions.
 
-## Instructions
-1. Analyze current build configuration
-2. Identify slow steps
-3. Implement caching strategies
-4. Verify improvements
-```
+## Why CLI-First?
 
-### Memory = Cross-Session State
+AI agents already live in the terminal. Wrapping them in a web UI or
+Python runtime adds latency, complexity, and failure modes. A CLI
+orchestrating CLIs is zero-overhead — and it means your agents can use
+**any tool you can run in a shell**.
 
-```bash
-# Agents accumulate knowledge
-squads memory show engineering
-# → "Switched to pnpm for faster installs"
-# → "Build cache reduced CI time by 40%"
+The more CLIs your agents have access to, the more capable your squads
+become. Squads itself is just the orchestrator — the real power comes
+from the tools you give your agents.
 
-# Search across all squads
-squads memory query "performance"
-```
+### Required
 
-## Commands
+| Tool | Purpose |
+|------|---------|
+| [Node.js](https://nodejs.org) >= 18 | Runtime |
+| [Git](https://git-scm.com) | Memory sync, version control |
+| [Claude Code](https://docs.anthropic.com/en/docs/claude-code) (`claude`) | Default agent execution |
 
-### Status & Dashboard
+### Recommended
 
-```bash
-squads status              # All squads overview
-squads status engineering  # Single squad details
-squads status -v           # Verbose with agent list
-squads dash                # Full dashboard with goals
-```
+| Tool | Purpose |
+|------|---------|
+| [GitHub CLI](https://cli.github.com) (`gh`) | Issue tracking, PRs, project management |
+| [Google Cloud CLI](https://cloud.google.com/sdk) (`gcloud`) | GCP deployments, secrets, infrastructure |
+| [Wrangler](https://developers.cloudflare.com/workers/wrangler/) (`wrangler`) | Cloudflare Workers, Pages, DNS |
+| [Google Workspace CLI](https://github.com/nicholasgasior/gws) (`gws`) | Drive, Gmail, Calendar, Sheets |
 
-### Running Agents
+### Any CLI Your Agents Need
 
 ```bash
-squads run engineering              # Run the whole squad
-squads run engineering/ci-optimizer # Run specific agent
-squads run engineering --dry-run    # Preview what would run
+terraform, kubectl, docker, aws, vercel, stripe, twilio,
+psql, redis-cli, curl, jq, ffmpeg, imagemagick...
 ```
 
-### Memory Management
+If it runs in a terminal, your agents can use it.
 
-```bash
-squads memory query "deployment"     # Semantic search
-squads memory show research          # View squad memory
-squads memory list                   # List all entries
-squads memory sync                   # Sync from git remote
-```
+### Skills + CLIs
 
-### Goal Tracking
+Agents get capabilities through two layers:
+
+- **CLIs** are the tools — `gh`, `gcloud`, `curl`, `psql`. They execute
+  actions in the real world.
+- **Skills** are the knowledge — markdown files that teach agents *how*
+  to use those tools effectively. A BigQuery skill teaches query
+  optimization patterns. A GitHub skill teaches your PR workflow. A
+  deployment skill codifies your staging-to-prod pipeline.
 
-```bash
-squads goal set finance "Cut costs 20%"  # Set goal
-squads goal list                          # View all goals
-squads goal progress finance 1 75         # Update progress
-squads goal complete finance 1            # Mark done
+```
+.claude/skills/
+├── bq/SKILL.md              # BigQuery patterns + cost optimization
+├── gh/SKILL.md              # GitHub workflow + PR conventions
+├── gcloud/SKILL.md          # GCP deployment procedures
+└── e2e-test/SKILL.md        # Browser testing with Chrome CDP
 ```
 
-### Feedback Loop
+Skills are injected into agent context alongside squad identity and
+memory. The combination of CLI tools + domain knowledge in skills is
+what turns a generic LLM into a specialized operator.
 
-```bash
-squads feedback add research 4 "Good analysis"   # Rate 1-5
-squads feedback show research                     # View history
-squads feedback stats                             # Summary
-```
+No MCP servers, no custom tool registries, no adapter layers. A skill
+file + a CLI installed on your machine is all an agent needs to operate
+in any domain.
 
-### Stack Management
+`squads doctor` checks which CLIs are available on your machine.
 
-```bash
-squads stack status        # Container health
-squads stack up            # Start Docker stack
-squads stack down          # Stop Docker stack
-squads stack health        # Comprehensive diagnostics
-squads stack logs bridge   # View container logs
-```
+## Commands
 
-### Updates
+The command surface is split into two audiences. Human operators manage
+the workforce — they set goals, monitor progress, and control budgets.
+Agents consume the CLI programmatically during execution — they read
+their own context, persist learnings, and record metrics. Every command
+supports `--json` so agents can parse outputs reliably.
+
+### For Humans
 
 ```bash
-squads update              # Interactive update
-squads update -y           # Auto-confirm
-squads update -c           # Check only
+# Setup
+squads init                    # Bootstrap .agents/ directory
+squads add <name>              # Add a new squad
+squads doctor                  # Check tools and readiness
+
+# Execute
+squads run <squad/agent>       # Run an agent or full squad
+squads autopilot               # Autonomous scheduling with budget control
+
+# Monitor
+squads status [squad]          # Overview of all squads
+squads sessions                # Active agent sessions on your machine
+squads dash                    # Dashboard with goals, metrics, activity
+
+# Goals & Tracking
+squads goal set squad "goal"   # Set a squad objective
+squads goal list               # View all goals
+squads results [squad]         # Git activity + KPI actuals
+squads stats [squad]           # Workforce scorecard + ROI
 ```
 
-## Claude Code Integration
+### For Agents
+
+Agents are the primary consumers of this CLI. After `squads run`
+dispatches an agent, it uses these commands to understand its context,
+persist knowledge, and evaluate its own work.
 
-### Option 1: Session Hook (Recommended)
+```bash
+# Context
+squads env show <squad> --json # Execution context (MCP, model, budget)
+squads env prompt <squad> -a <agent>  # Generate sub-agent prompt
+squads status --json           # Org-wide state for coordination
 
-Add to `.claude/settings.json`:
+# Memory
+squads memory read <squad>     # Recall squad knowledge
+squads memory write <squad> "x"# Persist a learning
+squads memory query "topic"    # Search across all memory
 
-```json
-{
-  "hooks": {
-    "SessionStart": [{
-      "hooks": [{
-        "type": "command",
-        "command": "squads status",
-        "timeout": 10
-      }]
-    }]
-  }
-}
+# Feedback loop
+squads feedback show <squad>   # Last cycle evaluation
+squads feedback add <squad> <rating> "text"  # Write evaluation
+squads exec list               # Own execution history
+squads kpi record <squad> <kpi> <value>  # Record a metric
 ```
 
-Now every Claude Code session starts with squad context.
+Everything above works locally — no login, no cloud, no API.
+Every command supports `--json` for machine consumption.
 
-### Option 2: CLAUDE.md Instructions
+## Configuration
 
-```markdown
-## Squads Workflow
+Agents need API keys to execute. Squads reads secrets from a `.env` file
+in your project root — the same pattern used by most Node.js and Python
+projects. Each provider's CLI uses its own environment variable, so you
+only need keys for the providers you actually use.
+
+```bash
+# .env — never commit this file
+ANTHROPIC_API_KEY=sk-ant-...    # Required for Claude Code (default provider)
+GEMINI_API_KEY=...               # Required for Gemini CLI
+OPENAI_API_KEY=sk-...            # Required for Codex
+GITHUB_TOKEN=ghp_...             # Recommended for gh CLI operations
+```
+
+`squads run` loads `.env` automatically before dispatching any agent.
+If a required key is missing, the provider's CLI will report the error —
+Squads doesn't mask or intercept auth failures. Add `.env` to your
+`.gitignore` to keep secrets out of version control.
+
+## Providers
+
+Multi-provider support isn't just a feature — it's central to how squads
+work. Different models have different strengths, costs, and speed
+profiles. A squad can route its scanner to a fast, cheap model (Gemini
+Flash) for high-volume monitoring, its worker to a deep reasoning model
+(Claude Opus) for complex analysis, and its verifier to a mid-tier model
+for cost-effective quality checks.
+
+Squads shells out to native AI CLIs. Each provider's CLI handles auth,
+context, and tool use independently — Squads just orchestrates.
+
+| Provider | CLI | Status |
+|----------|-----|--------|
+| Anthropic | `claude` | Stable — primary provider |
+| Google | `gemini` | Stable |
+| OpenAI | `codex` | Experimental |
+| Mistral | `vibe` | Experimental |
+| xAI | `grok` | Experimental |
+| Ollama | `ollama` | Experimental |
+
+Experimental providers have CLI integration but haven't been extensively
+tested in production. Contributions welcome — especially from teams
+already using these CLIs for autonomous work.
 
-Before starting work:
-1. Run `squads status` to see current state
-2. Run `squads memory query "<topic>"` to check existing knowledge
-3. After completing work, update memory via state files
+```bash
+squads run research --provider=google --model=gemini-2.5-flash
+squads providers    # List available providers and install status
 ```
 
-## Project Structure
+## Local Execution and Scaling
 
-```
-your-project/
-├── .agents/
-│   ├── squads/              # Squad definitions
-│   │   ├── engineering/
-│   │   │   ├── SQUAD.md     # Config + goals
-│   │   │   └── *.md         # Agent definitions
-│   │   └── research/
-│   ├── memory/              # Persistent state
-│   │   ├── engineering/
-│   │   │   └── state.md
-│   │   └── research/
-│   └── outputs/             # Agent outputs
-├── .claude/
-│   └── settings.json        # Hooks config
-└── CLAUDE.md                # Project instructions
-```
+Squads runs locally by default — your machine, your API keys, your
+control. There's no cloud dependency for core functionality. Each agent
+execution spawns a CLI process (`claude`, `gemini`, etc.) that runs
+until completion. Your data never leaves your machine unless the agent
+explicitly pushes to GitHub or another service you've configured.
 
-## Command Reference
+### Local limits
 
-```
-squads status [squad]         Show squad status
-  -v, --verbose               Include agent details
-
-squads dash                   Full dashboard with goals
-  -f, --full                  Include PRs and issues
-
-squads run <target>           Run squad or agent
-  -v, --verbose               Verbose output
-  -d, --dry-run               Preview only
-  -e, --execute               Execute via Claude CLI
-
-squads list                   List all squads/agents
-  -s, --squads                Squads only
-  -a, --agents                Agents only
-
-squads memory query <q>       Search memory
-  -s, --squad <squad>         Filter by squad
-squads memory show <squad>    View squad memory
-squads memory list            List all entries
-squads memory sync            Sync from git remote
-
-squads goal set <squad> <goal>
-squads goal list [squad]
-squads goal progress <squad> <idx> <pct>
-squads goal complete <squad> <idx>
-
-squads feedback add <squad> <rating> <text>
-squads feedback show <squad>
-squads feedback stats
-
-squads stack status           Container health
-squads stack up               Start Docker stack
-squads stack down             Stop Docker stack
-squads stack health           Comprehensive diagnostics
-squads stack logs <service>   View container logs
-
-squads update                 Interactive update
-  -y, --yes                   Auto-confirm
-  -c, --check                 Check only
-
-squads init                   Initialize project
-squads login/logout/whoami    Authentication (Pro)
-```
+| Parallel squads | Machine |
+|----------------|---------|
+| 2–3 | 8 GB RAM, 4 cores (laptop) |
+| 4–6 | 16 GB RAM, 8 cores (workstation) |
+| 8–12 | 32 GB+ RAM, 10+ cores (M-series Mac / desktop) |
+
+*Actual capacity depends on your CPU, memory, and which providers you
+use. `squads autopilot --max-parallel 3` controls concurrent executions.
+Monitor with `squads sessions`.*
+
+### Cloud scaling
+
+Local execution works well for individuals and small teams, but it has
+natural limits — your machine needs to stay running, parallel execution
+is bounded by hardware, and there's no shared visibility across team
+members. When you're ready to scale autonomous operations across teams,
+cloud execution runs the same agents, same memory, same commands — but
+on managed infrastructure instead of your laptop.
 
 ## Development
 
 ```bash
-git clone https://github.com/agents-squads/squads-cli
+git clone https://github.com/agents-squads/squads-cli.git
 cd squads-cli
 npm install
 npm run build
-npm link  # Test globally
+npm link       # Makes 'squads' available globally
+npm test
 ```
 
+TypeScript (strict mode), Commander.js, Vitest, tsup.
+
+## Contributing
+
+Contributions welcome. Open an issue first to discuss changes.
+
+1. Fork the repository
+2. Create your branch (`git checkout -b feature/my-feature`)
+3. Commit your changes
+4. Open a Pull Request
+
+See [CONTRIBUTING.md](CONTRIBUTING.md) for guidelines.
+
+## Community
+
+- [GitHub Issues](https://github.com/agents-squads/squads-cli/issues) — Bug reports and feature requests
+- [GitHub Discussions](https://github.com/agents-squads/squads-cli/discussions) — Questions and ideas
+- [Website](https://agents-squads.com) — Documentation and guides
+
 ## Related
 
-- [agents-squads](https://github.com/agents-squads/agents-squads) — Full framework with infrastructure
-- [engram](https://github.com/agents-squads/engram) — Persistent memory for AI agents
+- [agents-squads](https://github.com/agents-squads/agents-squads) — The framework
 
-## License
+## Get Started
 
-[MIT](LICENSE)
+```bash
+npm install -g squads-cli
+squads init
+squads run research/analyst
+```
 
----
+Build your own squads. Write your own skills. Optimize for your desired
+outputs. Every organization is different — Squads gives you the structure,
+you bring the domain knowledge. Start with the starter squads, observe
+what works, tune the context, and iterate. The agents get better as your
+skills and memory accumulate.
+
+We'd love to see what you build. Share your squads and skills in
+[GitHub Discussions](https://github.com/agents-squads/squads-cli/discussions).
 
-Built by [Agents Squads](https://agents-squads.com) — AI systems you can learn, understand, and trust.
+## License
+
+[MIT](LICENSE)