Quick Start

# Install
npm install -g specaegis

# Login (opens a browser flow)
specaegis login

# Start
specaegis

CLI Options

specaegis                     # Interactive mode
specaegis --ci <task>         # CI mode (non-interactive, JSON output)
specaegis --ci --resume-pipeline <run-id> # Explicitly resume a Pipeline run
specaegis login --device-code # Login from another device
specaegis --version           # Print CLI version

Commands

Command	Description
/pipeline <task>	Full pipeline: design when needed → code → quality gate → review → test
/design <task>	Technical design only; saves a design document without changing code
/test <desc>	Generate and run tests; only test files may be changed
/review	Review current git diff
/model	Interactive model picker (5 roles)
/connect	Manage BYOK sources (add/delete/test)
/status	Session status overlay
/compact	Manually trigger context compaction

Reference Files and Directories

Use @path in Chat and in every workflow command to provide existing workspace context. Quote paths with spaces using @"..."; add #L10-40 to select a text-file range.

Implement @docs/requirements.md and run the relevant tests.
Review @"docs/release notes.md#L10-40" for missing behavior.
/pipeline Read @docs/requirements/ and implement the selected feature.

Files are attached as bounded text. Directories provide only a non-recursive entry list, so the Agent reads relevant files on demand. References must stay in the current workspace; symlinks and outside paths are rejected. Type @ for file and directory completion, then use Tab or Enter to accept it.

Pipeline

Options

/pipeline <task>                    # Text description
/pipeline docs/spec.md              # Input file
/pipeline --dry-run <task>          # Preview without executing
/pipeline --skip-design <task>      # Start from code
/pipeline --from-design docs/design.md # Reuse a design document
/pipeline --retries 1 <task>        # Override retry limit
/pipeline --model qwen-max <task>   # Override model
/pipeline --yes <task>              # Skip plan confirmation

Stages

Stage	Role	Tools
Design	Architect	Read-only
Code	Engineer	All
Review	Reviewer	Read-only
Fix	Engineer + feedback	All
Test	QA	Tests only

Configuration

Use /config in the CLI to inspect active settings. User settings are stored in ~/.specaegis/settings.json; a project may provide .specaegis/settings.json.

BYOK (Bring Your Own Key)

Connect your own API keys to use models without consuming credits.

# Interactive panel
/connect

# Add, test, remove, or add models from the interactive panel

You can also manage BYOK sources in the Dashboard.

MCP Tools

Connect local stdio tools via Model Context Protocol. Create .specaegis/mcp.json:

{
  "servers": {
    "database": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres",
               "postgresql://localhost/mydb"]
    }
  }
}

API Reference

Create an Integration API Key in Dashboard → Integrations. It is separate from the credential used by the SpecAegis CLI.

LLM API Proxy

Use SpecAegis Gateway from compatible SDKs and your own applications with an Integration API Key.

OpenAI Format

# Python
from openai import OpenAI

client = OpenAI(
    base_url="https://api.specaegis.com/v1",
    api_key="YOUR_INTEGRATION_API_KEY"
)

response = client.chat.completions.create(
    model="qwen-max",
    messages=[{"role": "user", "content": "hello"}]
)

Anthropic Format

# Python
import anthropic

client = anthropic.Anthropic(
    base_url="https://api.specaegis.com/anthropic",
    api_key="YOUR_INTEGRATION_API_KEY"
)

message = client.messages.create(
    model="claude-sonnet-4.6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "hello"}]
)

Proxy Endpoints

Method	Path	Format
POST	https://api.specaegis.com/v1/chat/completions	OpenAI
GET	https://api.specaegis.com/v1/models	OpenAI
POST	https://api.specaegis.com/anthropic/v1/messages	Anthropic