Codex 3 — CapabilityOS
CapabilityOS: the shared operating system every division consumes and none forks.
- Authority rank
- 3
- Version
- v1.0
- Adopted
- unrecorded
- Held by
- Executive Office
- System
- SYS-02
Source · docs/codices/CODEX-3-capabilityos.md
CapabilityOS is the central operating system of Anabasis. Every division consumes it; no division forks it. It is owned by the Technology division and governed by this Codex.
Architectural rule
CapabilityOS is a set of bounded engines. Each engine declares its inputs, its outputs, its owning division, and the entities it may read. Engines communicate through declared interfaces, never by reaching into each other's storage. A new engine is added only when no existing engine can absorb the responsibility.
The engines
| Engine | Purpose | Reads | Writes |
|---|---|---|---|
| Knowledge Graph | The connective tissue: concepts, capabilities, sources, artifacts and their relations | all published knowledge | graph nodes and edges |
| Learning | Curriculum delivery, progress, assessment, credentialing | curriculum, enrolment, submissions | progress, assessments, credentials |
| Adaptive | Sequences what a person should do next based on demonstrated capability | assessments, goals, graph | recommendations |
| Goal | Long-horizon objectives and their decomposition | person, capability model | goals, milestones |
| Habit | Practice cadence, without manipulation or streak pressure | goals, activity | practice records |
| Decision | Structured decision support with recorded reasoning | goals, engagement context | decision records |
| Financial | Personal and institutional financial modelling | financial inputs | models, projections |
| Business | Enterprise diagnostics and operating models | organization data | diagnostics |
| Research | Publication pipeline, indexes, methodology | sources, outcome data | publications, index values |
| Advisory | Engagement lifecycle and delivery | engagements, advisors | engagement records |
| Venture | Founder and venture development pipeline | ventures, founders | venture records, investor reporting |
Dashboards
A dashboard is defined by the questions it answers, not the widgets it contains. Each is a view over engines; none owns data.
- Personal — Where am I becoming more capable, and what is next?
- Scholar — What have I completed, what have I demonstrated, what remains?
- Mentor — Who needs me, and where are they stuck?
- Advisor — What are my engagements, commitments, and open risks?
- Researcher — What is in the pipeline, and what does the evidence say?
- Founder — What stage am I at, and what is blocking the next one?
- Investor — What is the portfolio doing, and what has been reported?
- Enterprise — Is my organization becoming more capable, and where?
- Steward — Is the institution keeping its promises?
- Administrator — What is the operational state of the institution?
AI agents
Each agent has a scope, a permitted data set, a required human checkpoint, and an explainability requirement. Agents never act on consequential matters without a person confirming.
| Agent | Scope | Human checkpoint |
|---|---|---|
| Advisor Agent | Personal capability guidance | any commitment or financial action |
| Tutor Agent | Lesson-level explanation and practice | assessment and credentialing remain human-graded where they carry weight |
| Research Assistant | Source retrieval, summarization, drafting | publication requires editorial approval |
| Reviewer Agent | First-pass structural feedback on submissions | final review is peer or faculty |
| Analyst Agent | Institutional and enterprise analysis | interpretation and decision remain human |
Rules: no agent may read data the person could not read themselves; no agent output is presented as fact without its basis; no agent trains on member data without explicit, revocable, purpose-specific consent.