Part of #374
Question
What is the one command surface for seeing and controlling the graph, and how is the same surface exposed to the agent?
Starting sketch (to be grilled, not adopted)
agent-context-graph status
agent-context-graph sessions list|show|forget
agent-context-graph memories list|search|edit|forget
agent-context-graph entities show|forget
agent-context-graph capture pause|resume|exclude <path>
agent-context-graph reconcile status|retry
Sub-questions
- One CLI or per package?
sessions-graph reconcile exists separately today. Does the hub CLI own every noun and dispatch into packages the way hook does, or does each package keep its own CLI?
- Agent parity. Exposed through an MCP server, the plugin skill, or both? Can the agent run
forget without a human confirming it (this touches the no-human-gates rule; what automatic guard replaces it)?
- Provenance in views. Every
show should answer "where did this come from" (session → action/memory → chunk). Is that a flag or the default?
- Scope argument. Every command takes a scope (user now, shared later) so the multiplayer map doesn't force a redesign. What's the default?
- Output formats: human table plus
--json for the agent?
Part of #374
Question
What is the one command surface for seeing and controlling the graph, and how is the same surface exposed to the agent?
Starting sketch (to be grilled, not adopted)
Sub-questions
sessions-graph reconcileexists separately today. Does the hub CLI own every noun and dispatch into packages the wayhookdoes, or does each package keep its own CLI?forgetwithout a human confirming it (this touches the no-human-gates rule; what automatic guard replaces it)?showshould answer "where did this come from" (session → action/memory → chunk). Is that a flag or the default?--jsonfor the agent?