Starlight IntelligenceStarlightAcademy

A place to turn curiosity into craft

What would you like
to make possible?

Start with something that matters to you. Make a first attempt, question it, and leave with work you can keep improving.

Find my starting point ↓Explore the living Academy ↗

Free practice. No account or installation needed to begin.

Your first useful artifact · Example

A workflow you can explain.

The work
Prepare a weekly customer-feedback digest from a public sample.
AI prepares
A draft with links to the original comments.
You decide
Which conclusions hold up, and whether anything is shared.
Stop when
A source is missing or a comment contains private information.
Build your own workflow brief →Illustrative draft · hosted evaluation not run · no receipt issued

Choose by the work you want to do

One starting point.
Room to go deeper.

New to the Academy? Begin with a workflow. Each path opens a guided practice with sources, an artifact, review criteria and a way to return to your draft.

01 / Build a reliable workflow

Start here if you are unsure

Give your AI workflow a clear boundary

You have a recurring task and want to decide what AI may do.

Bring
One recurring task, its owner and an example input. No coding needed.
Make
A workflow brief with permitted actions, a review gate and a stop condition.

Your first move Name one task and the person who owns its result.

Begin this practice →Use my agent ↗
See the artifact and review criteriaAGENT_SYSTEM_CONSTITUTION.md
  • Bounded purpose. One owner, input, output, and stop condition are unambiguous.
  • Human authority. External and irreversible actions require named human approval.
  • Evidence quality. Sources, checks, exceptions, and uncertainty remain inspectable.
  • Recovery. A stop, escalation, and rollback path exist before operation.

Criteria guide your review. Completing a form does not establish mastery.

02 / Strengthen your judgment

Strengthen your judgment

Use AI without handing over your judgment

You write, create or make decisions with AI and want your thinking to stay visible.

Bring
One decision or creative task you can describe without private details.
Make
A personal pact for when to use AI, when to think independently and how to check yourself.

Your first move Choose a moment when you are tempted to accept an answer too quickly.

Begin this practice →Use my agent ↗
See the artifact and review criteriaAUGMENTATION_PACT.md
  • Independent judgment. An unaided view exists and the final decision is explained in the learner's own reasoning.
  • Epistemic quality. Claims, confidence, sources, and changed beliefs are distinguished.
  • Constructive challenge. AI is used to surface alternatives and tests rather than replace the decision.
  • Transfer. The learner demonstrates the reasoning pattern again without AI assistance.

Criteria guide your review. Completing a form does not establish mastery.

03 / Follow the evidence

Follow the evidence

Turn a fascinating claim into a testable question

You want to practice research that distinguishes observation from interpretation.

Bring
Curiosity about exoplanets. The mission links the scientific source material.
Make
An evidence map separating measurements, models and uncertainty about a distant world.

Your first move Write what you think we can know about a planet we cannot visit.

Begin this practice →Use my agent ↗
See the artifact and review criteriaEXOPLANET_EVIDENCE_MAP.md
  • Observation discipline. Measured signals are distinct from models, interpretations, and illustrations.
  • Source lineage. Each material claim points to an inspectable scientific source.
  • Uncertainty. Unknowns and plausible alternatives are named without false precision.
  • Transfer. The evidence pattern is correctly applied to another indirect signal.

Criteria guide your review. Completing a form does not establish mastery.

04 / Design an agent team

Design an agent team

Design a small team of agents you can inspect

You already understand a workflow and need explicit roles, handoffs and acceptance checks.

Bring
A bounded workflow and a named human owner. Some familiarity with agents helps.
Make
A department packet with roles, an operating loop, review criteria and human decisions.

Your first move Describe the department's one responsibility and the decisions it cannot make.

Begin this practice →Use my agent ↗
See the artifact and review criteriaAI_DEPARTMENT_PACKET.json
  • Workflow fit. The packet can recommend not automating and bounds one owner, input, output, exception, and reversible test.
  • Authority and risk. Human decisions, data limits, prohibited actions, stop conditions, and revocation are explicit.
  • Architecture simplicity. The smallest viable topology is justified and producer/verifier separation is preserved.
  • Evidence and evaluation. Completion evidence, representative tests, adversarial cases, uncertainty, and runtime pinning are inspectable.
  • Portability and recovery. Semantic components, native projections, adapter gaps, rollback, and revalidation triggers are named truthfully.

Criteria guide your review. Completing a form does not establish mastery.

A practice you can return to

The draft is
only the beginning.

Keep the moments that change your mind. Mission Studio holds the work in this page; download your practice file to keep it, then import it to continue.

Open the studio or import a draft →
  1. 01

    Think before you ask

    Describe the situation and preserve your own first attempt.

  2. 02

    Make something you can question

    Read the sources, build an artifact and attach critique to that exact draft.

  3. 03

    Revise. Then change the context.

    Explain what changed and try your method on a different situation.

The Academy, wherever you work

Your agent can
learn the method, too.

Install the free toolkit to bring mission coaching, department design and evaluation exercises into your own agent environment. The same learning contracts travel with you.

Install the Academy toolkit ↗Explore the five included skills →

Codex and Claude Code have verified Git installation paths. Other supported tools use the downloadable pack and host-specific setup.

Illustrative scene: a creator carries their folio out of the Academy.
Your practice belongs with your work.

A wider world of learning

Many branches.
Shared roots.

Explore the Knowledge Tree: evidence, your first agent, teams, memory, science, art and humane leadership. Follow one question into an example, a source and a practice you can keep.

Explore the Knowledge Tree →Make your first agent brief ↗

Eight branches of introductory learning. No account, hosted tutor or claim of mastery.

An imagined monumental tree with inhabited libraries and observatories
Curiosity grows through questions, practice and care.
Prefer to read first?

Free field guides →

Practical introductions to boundaries, evaluation and agent skills.

Want a smaller exercise?

Question one answer →

Judge a synthetic case, then inspect the reference and its limits.

Looking for a different perspective?

Meet the teaching guides →

Explore the Academy's fictional guides and their practice methods.

For developers: public learning contracts and current limits

Local practice · no account. Browser drafts live in page memory until you download them. Learner declarations and critique authorship are self-reported. No theatrical claims of consciousness, automatic assessment or agent authority accompany these exercises.

Use public, synthetic or redacted material. A human decides every external action. The public tools do not run a hosted tutor, issue credentials or connect a community.

Mission catalogFirst mission packetAgent start contractInstallation contract