Manage context
Context management turns evidence in a Scope into reusable artifacts, then assembles the selected results into a temporary context for one Agent turn. Use this order when learning or operating the system:
A Scope manages Sources, Artifacts, and Candidates. Sources preserve evidence. Artifacts can be produced by processing evidence or by explicit creation and updates where supported. Proposals requiring review enter a Candidate and commit an immutable Revision after approval; other paths follow their type's commit rules.
Committed artifacts support context recall, Handoff, Skill export, or Prompt configuration. See How context is created and used for the individual paths.
This chain describes data and authorization flow. It does not include availability checks, troubleshooting, or recovery; those belong to deployment and operations, outside this directory's data lifecycle.
Choose an entry point by content type
| What you need | Content type or concept | Start here |
|---|---|---|
| Understand boundaries, roles, and deployment relationships first | Architecture and responsibilities | Architecture and responsibilities |
| Isolate projects and control access | Scope | Scopes and access |
| Capture raw evidence or references to external material | Source | Sources and capture, Ingest text files |
| Read revisions and choose a write workflow | Artifact | Artifacts |
| Find relevant current artifacts and configure Topic or Atomic fusion | Artifact search | Search Artifacts, Fusion algorithms and parameters |
| Save and recall durable facts, decisions, and constraints | Memory | Memory and context |
| Build a reusable lesson and review it | Experience | Create and review Experience, Experience lifecycle |
| Manage an exportable instruction set | Skill | Create and export Skill, Configure Skill targets |
| Maintain stable Scope background information | Profile | Use Profiles |
| Derive topic summaries from long-running Source processing | Topic Memory | Use Topic Memory |
| Transfer task boundaries and outcomes across sessions | Handoff | Memory and Handoff, Handoff with Codex, Handoff Report |
| Add filterable metadata to artifacts | Tags | Manage Artifact tags |
| Configure Scope-level operational prompts | Prompt | Manage Prompts |
| Select and render context for one Agent turn | PreparedContext | Prepare context text |
| Enable optional vector or hybrid retrieval | Retrieval deployment option | Configure vector search |
The implementation currently writes these Artifact families: atomic-memory, topic-memory, experience, skill,
handoff, profile, and prompt. The memory family retains frozen legacy collection reads. A Source is evidence, not an Artifact; a Tag
is metadata, not an Artifact family. Confirm the actual surface in integration capabilities.
These guides describe the current implementation. Do not assume an Agent exposes every operation available through HTTP.