atlas

Atlas is a self-configuring multi-agent coding architect. Boots on /atlas and SessionStart: 18-agent squad, 8 automation hooks plus the atlas-doctor rollback guard (incl. the dispatch tripwire, which auto-flags orchestration runs so subagent discipline always engages), atlas-engine methodology (research -> test -> validate -> implement -> verify; evidence before any done claim), and 18 verification-gated /atlas-* launchers (incl. atlas-stacks for interactive skill stacking, atlas-launch, which opens a remediation session from an audit hub, and atlas-doctor, which diagnoses and repairs the plugin installation itself - including stale renamed-skill ghosts and broken orchestration wiring). Nine skills: atlas-engine (orchestration), atlas-stacks (AskUserQuestion-driven skill stacking), atlas-architect (boot/discovery), atlas-cartographer (architecture map + structural dedup), atlas-sextant (measurable self-improvement via observability DB), atlas-orbit (recurring loops), atlas-harbor (vendor MCP connectors), atlas-expedition (app-discovering UX swarm), atlas-survey (comprehensive quality/security/OWASP audit). Skills elicit user intent with AskUserQuestion when goal/scope/acceptance are genuinely the user's call. Integrates claude-mem and context-mode; recommends skills/plugins/MCPs the project needs.