Skip to content

Latest commit

 

History

History
33 lines (26 loc) · 2.02 KB

File metadata and controls

33 lines (26 loc) · 2.02 KB

Status: v0 alignment snapshot | 2026-08-11 | updated during implementation

Scope

In

  • All four layers: docket-core (work-queue engine) · docket-mcp (MCP surface) · docket-cc (Claude Code adapter) · docket-console (admin UI)
  • The worker/topic/item/claim domain model (core)
  • Item lifecycle: open → claimed → resolved → closed + resolution (architecture.md)
  • Stall detection (distinguishing "no owner" from "no update")
  • Every admin-console operation — especially "refine" (clean up an ambiguous request and re-queue it)
  • question — a stateless, fail-immediately request type
  • Single repo, fully public (ADR-0005)

Out

Item Rationale
File sync git already solves this. The core only deals with references (refs)
Automatic work distribution (orchestration) P-3; a human's manual assignment is the exception
Real-time chat Explicitly given up in the differentiation hypothesis (an axis it's fine to lose on)
Multi-worker collaborative claims Settled as single-claim only — exclusive claiming is part of the domain model's definition
Prompt-injection defense (in full) Excluded from v1 coverage — deferred to L2~L3
Regulatory compliance, performance tuning Settled as no hard constraints (principles.md)

Later

Item Condition Related doc
Multi-user (team-scale) Revisit once the auth approach is settled ADR-0006
Other agent runtimes (e.g. aims) The structure already supports adding just layer 3. Once real demand shows up architecture.md extension points
Human workers (a person picking up items from a phone) Possible with layer-4-only extension, no core changes. Once demand shows up architecture.md
Standardizing non-repo topic namespaces Once topics that aren't repos (org knowledge, machines, environments) actually start being used still undecided