Skip to content

concepts.md

Latest commit

 

History

History
94 lines (71 loc) · 3.42 KB

concepts.md

File metadata and controls

94 lines (71 loc) · 3.42 KB
Raw

Concepts

DoxieOS is built around a small set of durable concepts. The UI can evolve, but these concepts define the product model.

Core Objects

Concept What It Means Stored In
Agent A runnable Doxie skill with metadata, body, inputs, requirements, and output expectations. .doxie/skills/<id>/
Workflow A graph of agent and connector nodes that turns repeatable work into an inspectable run. .doxie/workflows/<id>/workflow.json
Library A reusable memory pack mounted into workspaces or dispatched agents. .doxie/libraries/<id>/
Workspace A persistent project folder where agents can read context and write artifacts. .workspace/ by default
Catalog A root folder containing Doxie skills, workflows, and libraries. .doxie/ or custom catalog folders
Provider A CLI-backed execution target such as Claude Code or OpenAI Codex. App settings and provider config
Run A persisted execution record for an agent or workflow. SQLite state and run output folders

Canonical Versus Generated

.doxie/ is the source of truth. Provider-specific folders are compatibility outputs so current CLIs can consume Doxie-authored assets.

flowchart TD
    Canonical[".doxie/ canonical catalog"] --> Regenerator["Doxie regenerator"]
    Regenerator --> ClaudeFiles[".claude/ generated compatibility"]
    Regenerator --> CodexFiles[".codex/ generated compatibility"]
    Canonical --> UI["DoxieOS UI"]
    Canonical --> Runner["Agent and workflow runner"]
Loading

Agents

An agent is the smallest reusable unit of Doxie work. It should define:

  • what the agent is responsible for;
  • what inputs it accepts;
  • which tools or external systems it needs;
  • what output another human or workflow node can consume;
  • when it should stop and ask for help.

Agents can be advisory, implementation-oriented, operational, workflow primitive, or handoff-focused.

Workflows

A workflow is a graph. It can run manually, on cron, or from persisted trigger definitions that will be wired later.

flowchart LR
    Trigger["Manual or Cron Trigger"] --> Intake["Intake Node"]
    Intake --> Analysis["Analysis Agent"]
    Analysis --> Decision{"Human Gate?"}
    Decision -->|Approve| Implement["Implementation Agent"]
    Decision -->|Reject| Stop["Cancelled Run"]
    Implement --> Review["Review Agent"]
    Review --> Output["Workspace Output"]
Loading

Workspaces

Workspaces make agent work less generic. They can hold project context, workspace memories, mounted libraries, generated outputs, and handoff artifacts.

When a workspace is selected, DoxieOS can inject:

  • WORKSPACE.md;
  • workspace memory/*.md;
  • mounted library memories;
  • operator hints collected during a run.

Libraries

Libraries are reusable memory packs. They are useful for standards, design systems, delivery playbooks, privacy rules, vendor integration notes, and team ways of working.

flowchart TD
    Library["Library"] --> Manifest["library.json"]
    Library --> Memories["memory markdown files"]
    Workspace["Workspace"] --> Mount["Mounted library"]
    Mount --> Dispatch["Agent dispatch context"]
    Memories --> Dispatch
Loading

Providers

DoxieOS is provider-neutral. It prepares the run context and dispatches to the selected provider CLI. Provider compatibility is intentionally an adapter layer, not the product model.

Supported provider families currently include Claude Code and OpenAI Codex.