- Five-minute governed workflow - install, create durable direction, review work, and verify a completed Task.
- User and Agent intake - setup, private local access, the organization dashboard, structured user tasks, and A2A intake.
- User approval policy - actions that require approval and the fail-closed baseline.
- Approval control - the owner-only exact-effect decision boundary.
- Completion review - user review of model-backed completion candidates.
- Planning and Task DAGs - bounded planning, atomic graph admission, dependency evidence, and external root isolation.
- Shared coordination - replayable same-Work peer Task context for Agent executions.
- Governed Lab - contained experiments, unverified results, and nomination-only promotion candidates.
- Organizational knowledge - evidence-backed versioned learning, lifecycle controls, and retrieval boundaries.
- Governance inspection - deterministic tenant-scoped runtime findings over chain-verified Event Contracts.
- SQLite recovery - backup, verification, restore, and rollback.
- Effect reconciliation - evidence-based recovery of uncertain external effects.
- AI management system - ISO/IEC 42001 readiness and the boundary between product controls and organizational conformity.
- A2A interoperability - the supported A2A profile and authorization boundary.
- Codex subscription provider - confined subscription-provider setup.
- OpenAI API provider - Responses API setup and approval requirements.
- Structured model input - instruction roles, invocation-bound source evidence, and transport limitations.
- Inference connections - exact account identity, shared budgets, atomic policy updates, and remaining runtime integration.
- Threat model - protected assets, attack surfaces, controls, and residual risks.
- Execution authority - supply-chain boundaries, consequential tool capabilities, containment prerequisites, and staged promotion.
- Event ledger integrity - stored-byte chaining, verification boundaries, and residual tampering risks.
- JSON boundaries - strict decoding and bounded untrusted input.
- Incident replay - bounded, payload-free reconstruction of one durable Work conversation.
- Release artifacts - reproducible Linux packages, corresponding source, and publication controls.
- Security policy - private vulnerability reporting.
Engineering contracts, acceptance evidence, and release gates are grouped under development.
Only this index and the threat model live directly under docs/. Documents use
lowercase filenames with dashes, except conventional names such as README.md.
| Folder | Contents |
|---|---|
guides/ |
User workflows and decisions |
architecture/ |
Runtime contracts and organizational state |
integrations/ |
Protocols, providers, and model inputs |
security/ |
Authority, integrity, and input controls |
operations/ |
Recovery, inspection, and release procedures |
governance/ |
Product governance controls and claim boundaries |
development/ |
Engineering contracts, checks, and technical readiness |
images/ |
Documentation images |
Implemented wire schemas live in the repository-root schemas directory.