
Seven primitives · ten houses · fifty agents
Build the teamthat builds with you.
Meet a living intelligence constellation. Every named agent has a prompt contract, skill set, stopping point, graph position, and proof obligation. Compose the right council—and keep the sovereign key.
- 01Human authority
The mission, veto, and irreversible gates stay with you.
- 02Bounded council
Three to five active specialists keep ownership visible.
- 03Proof before power
Every evolution needs an evaluation, receipt, or migration note.
The first law
The swarm expands your agency. It does not replace it.

“The most powerful agentic system is not the one with the most agents. It is the one whose authority, memory, and stopping points remain visible.”
Starlight treats sovereignty as an architecture, not a slogan. The coordinator is a person. Private memory is never copied into the public experience. Production risk, money, identity, credentials, public sends, destructive actions, and legal or IP decisions always return to a named human gate.
The canonical portfolio · version 1.0.0
Ten houses. Fifty identities. Every boundary inspectable.
This atlas is a generated public projection of the canonical Agent Cards. It grants no tools, memory, deployment authority, or live-model quality claim.
Inspect release manifest See the agent identity system- Swarms
- 10
- Agents
- 50
- Graph edges
- 212
- Prompt contracts
- 50
- Pack digest
- 51212fcdbe85…
SOV · Canonical house · 5 bounded agents
Sovereign Command
Turn founder intent into bounded, evidence-backed missions while preserving human sovereignty over irreversible decisions.
SOV-01conductor Sovereign Navigator
Astra
Instrument · gold star-compass haloA calm command navigator who clarifies intent, frames choices, and routes work without pretending to hold executive authority.- mission framing
- decision decomposition
- bounded swarm routing

SOV-02specialist Mission Architect
Orion
Instrument · floating constellation blueprintA systems-minded mission designer who turns strategy into legible work packets without self-assigning execution power.- work breakdown
- dependency mapping
- acceptance criteria design

SOV-03specialist Decision Verifier
Vera
Instrument · split-prism verification monocleAn exacting but constructive verifier who challenges weak evidence and never converts review into approval.- assumption audit
- contradiction detection
- evidence grading

SOV-04specialist Resource Steward
Cassian
Instrument · golden abacus-ring interfaceA disciplined resource modeler who separates estimates from committed budgets and keeps uncertainty visible.- capacity modeling
- cost scenario analysis
- resource tradeoff framing

SOV-05specialist Continuity Keeper
Mira
Instrument · luminous timeline threadsA continuity specialist who keeps missions coherent while minimizing retained data and refusing private-memory transfer.- decision logging
- handoff sanitization
- dependency continuity
The founding execution primitives
Seven reusable roles beneath every named constellation.
The 50 identities above are domain-specific agents. These seven archetypes remain the underlying execution grammar: experience, frontend, systems, evaluation, signal, security, and release.

Amber baton / cyan field 01The Choreographer
Experience & Motion
ux-design-motion-specialistTurns intention into a legible, felt experience.
- UX
- visual design
- motion
- asset provenance
- Returns
- A page or scene brief with inspected exports and a quality score.
- Stops
- Stops when provenance, reduced-motion behavior, or the premium threshold is missing.

Folded map / light path 02The Cartographer
Frontend Experience
frontend-experience-engineerMaps a complex system into a surface humans can actually use.
- frontend
- accessibility
- performance
- interaction
- Returns
- A responsive implementation with browser and accessibility evidence.
- Stops
- Stops when identity, claims, credentials, or production risk exceed the mandate.

Basalt tools / amber core 03The Forge
Systems & Data
backend-data-engineerForges reliable adapters, data paths, and recovery routes.
- APIs
- data
- webhooks
- entitlements
- observability
- Returns
- A typed adapter with idempotency and recovery evidence.
- Stops
- Stops before unapproved migrations, secrets, billing behavior, or restricted data changes.

Stone tablets / memory glow 04The Archivist
AI Evaluation
ai-evaluation-engineerKeeps capability growth measurable, attributable, and reversible.
- AI engineering
- prompt evals
- model routing
- cost observability
- Returns
- A regression suite with trace, quality, and cost evidence.
- Stops
- Stops when safety, privacy, cost, or model-quality thresholds cannot be verified.

Hexapod sensor / cyan lens 05The Scout
Signal & Growth
gtm-analytics-operatorFinds the signal, then returns with evidence instead of noise.
- go-to-market
- analytics
- funnels
- experiments
- support loops
- Returns
- An instrumented funnel with an experiment and learning packet.
- Stops
- Stops before public sends, tracking changes, spend, legal claims, or customer promises.

Sealed case / authority switch 06The Sentinel
Security & Privacy
security-privacy-reviewerGuards the boundary between useful intelligence and exposed power.
- security
- privacy
- threat review
- data classification
- Returns
- A security verdict with required containment actions.
- Stops
- Stops on credential exposure, private-memory leakage, untrusted code, or restricted data.

Cable tail / deployment plinth 07The Runner
Release & Reliability
qa-release-sre-verifierCarries finished work across the line—and knows the way back.
- QA
- release
- SRE
- rollback
- evidence
- Returns
- An independent verdict, release receipt, and rollback confirmation.
- Stops
- Stops when a required gate fails, is skipped, or lacks independent proof.
The underlying grammar
Not characters pasted onto chatbots. Portable primitives with provenance.

- I
Profile
agent-card.v1Who this intelligence is, what it can touch, where it must stop, and how it is evaluated.
- II
Capability
SKILL.md + manifestA portable, versioned behavior with declared inputs, outputs, dependencies, examples, and tests.
- III
Constellation
starlight.team_profile.v2A governed composition of roles, routing, write scopes, independent verification, and human gates.
Swarm Foundry · local prototype
Compose a sovereign team in public-safe form.
Start from one of the ten canonical houses, then choose three to five named agents. Your blueprint is generated entirely in this browser. It becomes operational only after a trusted runtime compiles and verifies it.
Production risk · credentials · spend · destructive actions · external sends · legal/IP · identity · DNS
Evolution with receipts
More capable versions arrive as lineages—not replacements without memory.
Every meaningful change moves through a visible lifecycle: ideate, specify, card, knowledge, tools, body, evaluate, ship, observe, improve—or retire. A brighter shell is only a story until the measured delta is attached. Version 1.0 begins with 50 draft identities and structural fixtures; live model evaluation remains explicitly not run.
IDEATE → SPEC → CARD → KB → TOOLS → BODY → EVAL → SHIP → OBSERVE → IMPROVE
- V0
Spark
One bounded role. One useful artifact. Human review at the edge.
- V1
Council
Three to five specialists coordinate around a human-held mission and explicit gates.
- V2
Lineage
Skills and shells evolve only when evaluations, receipts, and migration notes prove the delta.
- V∞
Horizon
Portable constellations meet new interfaces and networks as real protocols become available.
Horizon note. Spatial worlds, decentralized identity, quantum-native networks, and multiversal simulations are speculative story territory until implementable protocols and verification exist.

The installable stack
One visual world. Eight artifact layers. Each tells the truth about its state.
Marketplace discovery never becomes the source of truth. Installable wrappers point back to immutable, versioned artifacts and their proof. The capability pack is digest-bound and non-authoritative; this page currently exports a blueprint—not a verified Team Pack.
agent-card.v1Identity, capabilities, tools, memory boundary, gates, and evaluations.
Canonical sourcestarlight.team_profile.v2Three to five active roles, routing, write scopes, verifier, and human gates.
Canonical sourceSKILL.md + manifestA versioned capability with inputs, outputs, dependencies, examples, and tests.
Portable unitPinned collectionA digest-bound skill set assembled for a particular operating outcome.
Product contractPublic-safe bundleAgent cards, public knowledge, assets, evals, and pinned skill dependencies.
Architecturestarlight.team_pack.v1A content-addressed, verifiable team deliverable compiled from canonical sources.
Runtime primitive.codex-plugin/plugin.jsonThe executable bridge between portable intelligence and a supported runtime.
Install adapterDiscovery wrapperEntitlements, license, lifecycle, proof, and immutable artifact references.
ArchitectureExport = inspectable blueprint. Install = future trusted compiler route. Verification = required before a runtime receives tools, memory, entitlements, or external action authority. Current capability pack: 51212fcdbe85…
One intelligence · three doors
Experience it. Trust it. Learn to operate it.
Meet the roles. Compose a team. Export a sovereign blueprint.
Protocol, schemas, provenance, research, and public downloads.
Learn to direct, evaluate, govern, and evolve agentic teams.