Specswarm

SpecSwarm v7.19.3

Spec-driven development for Claude Code. Build → Fix → Modify → Ship, with quality gates, multi-agent orchestration, version-controlled specs, autonomous chunk execution (v7.1.0–v7.10.0), and the AUTO-MAGIC loop (v7.13.0–v7.17.0): a taste model that learns your rulings, assume-first clarification, adversarial verification with teeth, and /ss:go — a full feature ladder with a median of ≤4 human touchpoints. v7.18.0 adds the mentor→builder conduct loop (/ss:conduct + /ss:mentor-init), extracted from ~500 audited production dispatches.

Install

Restart Claude Code to activate the plugin. *(Upgrading from v5.x? See Migrating from v5.x at the bottom.)*

The core commands

v7.1.0–v7.18.0 autonomous-loop commands

The core commands above remain the canonical loop (with /ss:go as the v7.17.0 default entry point). The v7.1+ commands compose with them to enable single-session execution and unattended chunks. Each is optional; the core loop works without any of them.

Together these implement the autonomous chunk loop: pre-batch decisions at 9pm, schedule /ss:overnight via cron between 10pm-6am, wake to a phone notification (success or "needs review"). The dual mentor↔builder session pattern is now optional, not required. See CHANGELOG.md for the full architectural story.

v7.19.0 — ss-status: the statusline is the status surface. A new separate plugin in this marketplace (/plugin install ss-status@specswarm-marketplace): a rich statusline plus a real-time background-builder segment for conduct projects — 🔨 building:   / 💀 last build DIED / ✓ builders idle — safe to close. One-time /ss-status:install wraps whatever statusline you already have (yours keeps rendering, cached; the segment refreshes every 5s even while idle). Born from a day-one v7.18.0 lesson:…