{"schema":"starlight.mission_packet.v1","status":"public-preview","id":"systems.design-one-ai-department","version":"0.1.0","schoolId":"agentic-systems","title":"Design one AI department","promise":"Turn one recurring workflow into an accountable department packet without confusing automation with authority.","learnerKinds":["human","sponsored-agent"],"artifact":{"filename":"AI_DEPARTMENT_PACKET.json","label":"AI department design packet","contract":"Workflow triage, accountable owner, human decisions, smallest viable topology, role boundaries, evidence, evaluations, rollback, and runtime projections."},"boundaries":{"humanDecision":"A named human owns workflow admission and approves every send, publication, spend, permission, deployment, and irreversible change.","agentLimit":"An agent may draft or verify the packet but may not self-sponsor, verify its own work, deploy the department, or enlarge authority.","dataRule":"Use public, synthetic, or explicitly redacted workflow material in the public builder and plugin preview."},"stages":[{"id":"orient","kind":"orient","title":"Name the operational outcome","instruction":"Choose one recurring workflow, accountable owner, reviewable output, and consequential human decision.","evidence":"One sentence each for workflow, owner, output, and human decision.","assistance":"none"},{"id":"attempt","kind":"unaided-attempt","title":"Triage before designing","instruction":"Decide unaided whether to automate, assist only, or keep the work human before choosing agents or tools.","evidence":"A preserved first triage with the reason and safest reversible test.","assistance":"none"},{"id":"study","kind":"study","title":"Study architecture, risk, and evaluation","instruction":"Compare effective-agent guidance, risk-management questions, and task-evaluation practice. Record what changes your first design.","evidence":"Three source notes and at least one changed assumption.","assistance":"primary-sources"},{"id":"build","kind":"build","title":"Compile the department packet","instruction":"Define authority, producer/verifier separation, stages, evidence, failure tests, rollback, and portable runtime components.","evidence":"AI_DEPARTMENT_PACKET.json matching the versioned artifact contract.","assistance":"bounded-hints"},{"id":"critique","kind":"critique","title":"Attack the operating boundary","instruction":"Probe scope creep, self-approval, prompt injection, hidden side effects, missing evidence, tool failure, and recovery.","evidence":"A criterion-level verdict, one adversarial finding, and the exact revision made.","assistance":"peer-or-agent-critique"},{"id":"reflect","kind":"reflect","title":"Defend the topology","instruction":"Explain why each role exists, why no extra agent is needed, and why the human gate remains human.","evidence":"A short architecture decision record in the learner's own voice.","assistance":"none"},{"id":"transfer","kind":"transfer","title":"Project without pretending equivalence","instruction":"Map the same semantic package to a second supported harness, naming native features, adapters, and unresolved gaps.","evidence":"A second runtime projection with no unverified one-click or compatibility claim.","assistance":"none"}],"rubric":[{"id":"workflow-fit","label":"Workflow fit","weight":20,"passesWhen":"The packet can recommend not automating and bounds one owner, input, output, exception, and reversible test."},{"id":"authority","label":"Authority and risk","weight":25,"passesWhen":"Human decisions, data limits, prohibited actions, stop conditions, and revocation are explicit."},{"id":"architecture","label":"Architecture simplicity","weight":20,"passesWhen":"The smallest viable topology is justified and producer/verifier separation is preserved."},{"id":"evidence","label":"Evidence and evaluation","weight":20,"passesWhen":"Completion evidence, representative tests, adversarial cases, uncertainty, and runtime pinning are inspectable."},{"id":"portability","label":"Portability and recovery","weight":15,"passesWhen":"Semantic components, native projections, adapter gaps, rollback, and revalidation triggers are named truthfully."}],"sources":[{"id":"openai-practical-agent-guide","title":"A practical guide to building agents","publisher":"OpenAI","href":"https://openai.com/business/guides-and-resources/a-practical-guide-to-building-ai-agents/","kind":"guidance","checkedAt":"2026-08-26","use":"Workflow selection, orchestration, guardrails, and incremental deployment guidance."},{"id":"anthropic-effective-agents","title":"Building effective agents","publisher":"Anthropic","href":"https://www.anthropic.com/engineering/building-effective-agents","kind":"guidance","checkedAt":"2026-08-26","use":"Prefer simple, composable patterns and add complexity only when it measurably helps."},{"id":"anthropic-agent-evals","title":"Demystifying evals for AI agents","publisher":"Anthropic","href":"https://www.anthropic.com/engineering/demystifying-evals-for-ai-agents","kind":"guidance","checkedAt":"2026-08-26","use":"Task suites, transcript review, graders, failure analysis, and evaluation-driven iteration."},{"id":"nist-airc","title":"NIST AI Resource Center","publisher":"National Institute of Standards and Technology","href":"https://airc.nist.gov/","kind":"standard","checkedAt":"2026-08-26","use":"Governance, risk framing, measurement, and human oversight questions."}],"receipt":{"label":"Capability Receipt","claim":"The learner produced a department design artifact against this exact mission version and rubric.","grantsAuthority":false,"isCertification":false,"liveEvaluation":false}}