specfront-ai (npm)

Interactive CLI client for the SpecFront AI agent server. Pure Node — zero Python dependency. Connects to a self-hosted or Azure-hosted SpecFront server over HTTP+SSE.

Install

npm install -g specfront-ai
# or
npx specfront-ai chat

Configure

Point the CLI at your SpecFront server:

export SPECFRONT_SERVER=http://localhost:8000        # local dev
# or
export SPECFRONT_SERVER=https://specfront.yourco.com # hosted

Optional ~/.specfront/config.toml:

server = "http://localhost:8000"
# token = "..."  # reserved

Use

Bare specfront enters the interactive REPL:

$ specfront
╭ SpecFront interactive chat ──────────────────────────────────────────╮
│ Server: http://localhost:8000                                         │
│ Type /help for commands, free text at a checkpoint = reject feedback. │
╰───────────────────────────────────────────────────────────────────────╯
✓ connected — {"status":"ok","service":"specfront-ai","version":"0.3.0"}
› /run --jira SCRUM-5 --mockup ./login.png --repo owner/repo
  …streaming agent updates…
⏸ HITL checkpoint reached.
› the logo should be inline SVG, not a placeholder img tag
↳ recorded feedback, resuming with rejection
  …
› /approve
✓ run complete

Or use subcommands:

specfront run --jira SCRUM-5 --mockup ./login.png --repo owner/repo
specfront resume <thread-id> --approve
specfront resume <thread-id> --reject "feedback"
specfront list
specfront history <thread-id>
specfront status <thread-id>
specfront health

Slash commands

/run, /approve, /reject, /show, /status, /history, /cost, /list, /thread, /fork, /cancel, /export, /models, /clear, /help, /exit. Free text at a checkpoint is treated as a rejection with that feedback.

Architecture

This package contains only an HTTP client. The actual agent (LangGraph, LangChain, model calls) runs in a separate server — see specfront_ai on GitHub.