feat: molt - self-healing update agent

[CoreyH]

Summary

Molt is a self-healing update system for self-hosted Clawdbot instances. It handles nightly updates from upstream with automatic rollback if something breaks.

Lobsters molt to grow — shedding their old shell and emerging fresh.

The Problem

Running Clawdbot on a self-hosted server means you want automatic updates. But updates can break things:

• pnpm install fails (network, deps)
• New code crashes the gateway
• Gateway doesn't come back up
• You're asleep when this happens

Currently: bot goes silent, you discover hours later, SSH in, diagnose, fix.

The Solution

Molt runs nightly and:

1. Updates safely — pulls, installs, builds, restarts
2. Verifies health — stability window catches crash loops
3. Rolls back automatically — if health check fails
4. Reports status — changelog + Slack notification

Key Design Decisions

Agentic Recovery

Traditional self-updaters are deterministic (blue/green, staging dirs). Molt is simpler: try, verify, rollback if broken. If rollback also fails, capture context for the AI (or human) to fix it.

Module Manifest

Personalized health checks. If you only use Slack, a broken Discord adapter isn't worth rolling back for:

{
  "modules": { "channels": ["slack"] },
  "healthCriteria": { "channels": "all" }
}

Files Changed

• docs/molt.md — Full PRD with architecture and rationale
• scripts/molt.sh — Reference implementation (bash)
• .gitignore — Ignore local workspace files

Status

Running in production. Successfully updated 289 commits this morning:

STATUS: SUCCESS
FROM: 309fcc53 → TO: 003fff06
COMMITS: 289

Open Questions

1. Where should this live? Core, plugin, or standalone?
2. Auto-discover module manifest from config?
3. TypeScript rewrite vs bash?

RFC — seeking feedback before formal upstream PR.