Experiment & Demo Layer

5 Labs โ€” Hands-On Agent Engineering

Small, explainable, reusable demos for each core capability. Each lab is designed to teach one pattern through practice.

๐Ÿ’ญ

Memory Lab

ready

Show how a project evolves when rules, build commands, and prior learnings become persistent assets.

Demo Flow
01Memory entry creation and 3-layer architecture demo
02Memory update and contradiction resolution
03Memory lookup and verification against codebase
04autoDream-style consolidation simulation
Prerequisites
โ€ข Understanding of MEMORY.md index pattern
โ€ข Familiarity with topic file structure
Governance Notes

Memory must be treated as hints, not truth. Always verify against actual project state before acting.

Try Interactive Demo โ†’Related capability โ†’
๐Ÿช

Hook Lab

ready

Show how standards become execution checkpoints rather than documentation that gets ignored.

Demo Flow
01Pre-edit warning hook (prevent changes to protected files)
02Post-edit lint/test hook (auto-run validation)
03Pre-publish validation hook (check V5.3 compliance)
04Policy block demo (reject non-compliant operations)
Prerequisites
โ€ข Understanding of lifecycle events
โ€ข Basic shell scripting
Governance Notes

Hooks enforce standards automatically. The leak was caused by a missing hook-equivalent check in the build pipeline.

Try Interactive Demo โ†’Related capability โ†’
โšก

Skill Lab

ready

Show how repeatable workflows are packaged and reused across projects and roles.

Demo Flow
01Create a reusable skill (SKILL.md format)
02Invoke skill from natural language
03Edit and version a skill
04Skill library management view
Prerequisites
โ€ข Understanding of skill packaging format
โ€ข Familiarity with prompt templates
Governance Notes

Skills should be reviewed before distribution. Version control and rollback capability required.

Try Interactive Demo โ†’Related capability โ†’
๐ŸŽฏ

Subagent Lab

ready

Show decomposition of complex tasks into specialist roles with clear accountability.

Demo Flow
01Fork mode: independent parallel research
02Teammate mode: collaborative writing with shared context
03Advisor pattern: quality oversight of worker output
04Multi-agent pipeline: research โ†’ write โ†’ review โ†’ deploy
Prerequisites
โ€ข Understanding of Fork/Teammate/Worktree models
โ€ข Familiarity with mailbox communication
Governance Notes

Explicit ban on lazy delegation. Coordinator must read actual findings, not just summaries.

Try Interactive Demo โ†’Related capability โ†’
๐ŸŒ™

NightOps Lab

ready

Demonstrate unattended nightly operational patterns that remain governable and auditable.

Demo Flow
01Nightly summary generation from daily logs
02Work queue processing with priority sorting
03Morning brief compilation
04Retry and recovery logic with failure budgets
Prerequisites
โ€ข Understanding of KAIROS daemon architecture
โ€ข Familiarity with append-only logging
Governance Notes

15-second blocking budget per action. Actions exceeding budget must be deferred. All operations append-only logged.

Try Interactive Demo โ†’Related capability โ†’
๐Ÿ—๏ธ

AI OS Lab: LightHope 10-Layer Architecture

planned

Show how a real-world AI OS is built with protocol-first design: 10 processing layers, skill registry, policy engine, and audit layer working together. Based on m8101 LightHope AI OS v1.3.0.

Demo Flow
0110-layer pipeline walkthrough: input โ†’ normalization โ†’ intent โ†’ task โ†’ policy โ†’ skills โ†’ context โ†’ model โ†’ audit โ†’ memory
02Skill Registry: register and invoke SK-008 problem_selector
03Policy Engine: YAML-driven adaptive difficulty adjustment
04Audit Layer: three-dimension output quality check (format/goal/risk)
05Memory stratification: session vs user vs project scope demo
06Dual-channel routing: local Ollama vs Claude API escalation
Prerequisites
โ€ข Understanding of layered architecture patterns
โ€ข Familiarity with Python FastAPI basics
Governance Notes

Seven Iron Laws (ไธƒ้“ๅพ‹): business rules in policy not if-else, capabilities via skill registry not hardcoded, model output audited independently, prompt does only current step. These laws are enforced at the architecture level, not the prompt level.

Related capability โ†’
๐Ÿ“š

Knowledge Pipeline Lab

planned

Demonstrate the m8100โ†’m590โ†’m495 knowledge pipeline: from raw source to compiled knowledge to exportable skill.

Demo Flow
01Source ingestion: parse a research paper into normalized format
02Knowledge compilation: agent-driven extraction of claims with evidence
03Quality gate evaluation: multi-dimensional scoring (trust, confidence, freshness)
04Skill export: transform compiled knowledge into m495 6-file skill package
05JSONL feed generation: create structured feed for m590 ingestion
06Conflict detection: handle contradicting sources without silent overwrite
Prerequisites
โ€ข Understanding of m590 six-layer architecture
โ€ข Familiarity with JSONL data format
Governance Notes

Knowledge pipeline must preserve provenance at every stage. Every compiled claim links back to its source with confidence score and trust level.

Related capability โ†’
๐Ÿช–

CSAR Doctrine Lab

planned

Show how military doctrine is encoded into structured knowledge objects that drive m595 COMPASS gameplay mechanics.

Demo Flow
01CSAR five-phase walkthrough: report โ†’ locate โ†’ authenticate โ†’ support โ†’ recover
02Authentication challenge: ISOPREP-based identity verification under time pressure
03Deception layer detection: identify misinformation, impersonation, and signal manipulation
04Evasion route planning: PACE methodology with terrain analysis
05Radio protocol simulation: frequency management and transmission risk assessment
06Coalition coordination: multi-national force communication challenges
Prerequisites
โ€ข Basic understanding of military operations concepts
โ€ข Familiarity with m595 COMPASS game design
Governance Notes

All CSAR content derived from public doctrine (JP 3-50, public training materials). No classified information. Legal boundary strictly maintained.

Related capability โ†’
๐ŸŽฎ

Game Design Mechanics Lab

planned

Demonstrate core game design patterns that make m595 COMPASS engaging: information asymmetry, moral weight, adaptive difficulty, and procedural narrative.

Demo Flow
01Information asymmetry: different actors see different information in the same scenario
02Moral weight decision: abort rescue to protect team vs. risk team to save pilot
03Adaptive difficulty: invisible challenge adjustment based on player performance
04Procedural narrative: generate unique missions from theater geography + force composition
05Time pressure: fuel-limited extraction with approaching enemy patrol
06Resource scarcity: allocate limited supplies across competing needs
Prerequisites
โ€ข Basic game design concepts
โ€ข Understanding of narrative mechanics
Governance Notes

Game mechanics must serve realism first, entertainment second. Realism score (100/100) is a prerequisite before fun score optimization.

Related capability โ†’
๐Ÿ”„

Pipeline Integration Lab

planned

Demonstrate the full five-project industrial pipeline: m8100 (search) โ†’ m590 (compile) โ†’ m495 (skill) โ†’ m553 (media) โ†’ m595 (game).

Demo Flow
01m8100 search: discover new CSAR authentication techniques from public sources
02m590 compile: transform raw research into structured knowledge objects
03m495 export: package knowledge as installable game skill
04m553 generate: create visual assets driven by m590 knowledge data
05m595 integrate: install skill and verify in game scenario
06End-to-end: trace a single knowledge seed from search to gameplay
Prerequisites
โ€ข Understanding of the five-project ecosystem
โ€ข Access to project APIs (m590, m495, m553, m595)
Governance Notes

Pipeline status levels: ACTIVATED (interfaces defined) โ†’ PROVEN (data flows end-to-end) โ†’ INDUSTRIAL (automated with monitoring). Each milestone upgrades status.

Related capability โ†’