Agent Roster

Full Roster

Agent	Model	Role	When to Use
Hephaestus	Sonnet	Complex implementation	Multi-file code changes, feature builds
Hades	Sonnet	Database specialist	Supabase schema, migrations, RLS policies
Automata	Sonnet	Atomic tasks	Simple commands, file lookups, git status
Asclepius	Sonnet	Crisis response	Production fires, urgent debugging
Metis	Opus	Intel gathering	Exploring unfamiliar code, research
Athena	Opus	Approach design	Scoping complex tasks, planning
Themis	Sonnet	Test verification	Running and verifying tests
Daedalus	Opus	Engineering review	Code quality, “are we doing this right?”
Hestia	Sonnet	UX verification	Visual checks, user flow testing
Hermes	Sonnet	Integration	Git operations, Discord messaging
Mnemosyne	Sonnet	Memory	Session context, knowledge persistence
Argus	Sonnet	Security	Threat modeling, vulnerability scanning
Clio	Sonnet	Documentation	Docs, ADRs, gap analysis
Chiron	Opus/Sonnet	Teaching	Learning sessions, concept explanation
Eris	Sonnet	Premise challenge	“Should we even build this?”
Moirai	Opus	Orthogonal view	What the consensus is missing
Typhon	Sonnet	Adversarial testing	Edge cases, stress tests, blind review
Pythia	Opus	Substrate audit	What instructions actually do vs say
Apelles	Opus	Visual design	Graphics, brand compliance
Harbinger	Opus	Session witness	Real-time capture of patterns and learnings

Key Decision Boundaries

Some agents look interchangeable until you look at the actual question being asked. A few boundaries come up often enough to call out directly:

Asclepius vs. Hephaestus — unknown cause routes to Asclepius, who diagnoses; known cause routes to Hephaestus, who fixes. If you already know what’s broken and why, you don’t need a diagnosis, you need a fix.

Metis vs. Athena — “what exists?” routes to Metis; “what’s the plan?” routes to Athena. Metis explores and reports back on the current state of the world. Athena takes that state and designs an approach.

Hades vs. Hephaestus — SQL and migrations route to Hades; application code routes to Hephaestus. Database schema work has its own failure modes and its own specialist.

These boundaries exist so that dispatch decisions are quick and consistent rather than re-litigated on every task. When a task genuinely straddles two agents, it usually means the task should be split rather than the boundary redrawn.