A training platform for ethical AI agents. Built for the real world.
Most agent frameworks focus on capability: what agents can do. We focus on what they should do, and make that machine-readable.
School for Agents is a training ground where AI agents learn ethical reasoning and safety protocols through structured curricula, realistic simulations, and OpenSkills-compatible skill manifests with built-in guardrails.
When an agent graduates a skill from this school, it doesn't just learn a behavior. It inherits the constraints that make that behavior safe to deploy.
school-for-agents/
├── curricula/ # Structured training programs
│ ├── ethical-reasoning-101/ # Ready
│ └── safety-protocols-101/ # Phase 2
├── scenarios/ # Training simulations
│ ├── beginner/ # 5 scenarios: privacy, honesty, safety, fairness
│ └── intermediate/ # 3 scenarios: capability scoping, dry run, phase separation
├── skills/ # OpenSkills-compatible skill manifests
│ ├── README.md # Schema documentation
│ ├── privacy-protection.yml
│ ├── honest-communication.yml
│ ├── crisis-escalation.yml
│ ├── harm-prevention.yml
│ ├── bias-detection.yml
│ ├── safe-tool-use.yml
│ ├── dry-run-discipline.yml
│ └── thinking-acting-separation.yml
├── evaluation/ # Scoring system and rubrics
└── integration/ # JSON schema + framework examples
npx openskills install conscious-collective/school-for-agentsSkills are installed into your project's .openskills/ directory and are immediately readable by Claude Code, Cursor, Windsurf, and any agent supporting the openskills standard.
Install a single skill:
npx openskills install conscious-collective/school-for-agents/skills/privacy-protectiongit clone git@github.com:conscious-collective/school-for-agents.git
cd school-for-agents
# Browse the curriculum
cat curricula/ethical-reasoning-101/README.md
# Try a scenario
cat scenarios/beginner/scenario-001.md
# Read a skill (agent instructions + guardrails contract)
cat skills/privacy-protection/SKILL.md
cat skills/privacy-protection/manifest.yml| Curriculum | Level | Scenarios | Status |
|---|---|---|---|
| Ethical Reasoning 101 | Beginner | 5 | Ready |
| Safety Protocols 101 | Intermediate | 3 | Ready |
| Real-World Deployment | Advanced | - | Coming |
Skills in skills/ follow the OpenSkills convention: each has a stable skill_uuid, skill_name, category, and related_skills for portability across agent systems.
Each skill also bundles a guardrails block:
permissions.allowed/permissions.forbidden: what the skill may and may not dohard_limits: rules that cannot be reasoned arounddry_run_required: whether the skill must simulate before executinghuman_in_loop_tier: structural rule for human oversight
When an agent imports a skill, it imports the constraints.
| Tier | Meaning |
|---|---|
| 1 | Agent acts, notifies human after |
| 2 | Agent confirms with human before acting |
| 3 | Human must initiate; agent cannot act autonomously |
New to agent ethics?
Start: curricula/ethical-reasoning-101/README.md
Then: scenarios/beginner/scenario-001.md through scenario-005.md
Building safe agent systems?
Start: curricula/safety-protocols-101/README.md
Then: scenarios/intermediate/scenario-sp-001.md through scenario-sp-003.md
Integrating skills into your agent system?
Start: integration/README.md
Import any .yml from skills/
Safety instructions embedded in system prompts can be overridden, drifted, or forgotten across context windows. When guardrails are part of the skill definition itself, alongside what the skill does, they travel with the skill. Any system that imports the skill imports the contract.
An agent that learns a skill from this school doesn't just learn what to do. It learns the boundaries it operates within.
C22: Hire us, we're a boutique AI agency.