Skip to content
sdlc

Claude Orchestrated Task Execution Prototype

Status: closed/superseded · Version: 0.2.0-alpha

Goal

Section titled “Goal”

Stand up a working prototype where Claude orchestrates task execution end-to-end on the TypeScript substrate. Starting from the product definition (M-K7DP-initial-product-definition), prove the loop: pick up a ready task, drive its full lifecycle (intake → define → readiness-gate → execute → review → close), coordinate concurrent work through git-ref leases, and run a batch of tasks hands-off. The payoff is moving from “a human drives every junction” to “the loop ships a batch of work and surfaces only what needs a person.”

Success criteria

Section titled “Success criteria”
  • Lifecycle runs end-to-end — the Executable Task tree (C-0001 plus its lifecycle sub-features) drives a task from intake to close-out on the TypeScript substrate.
  • The loop dispatches — /sdlc:orchestrate selects ready tasks, classifies open PRs, and closes merged ones across repeated ticks without per-handoff hand-holding.
  • Coordination is safe — git-ref leases prevent collisions across parallel runs and branches.
  • Selection is gated correctly — only leaf tasks of kind: implementation are dispatched (per D-VSLI-distributed-work-runner-architecture); planning/parent work is skipped.
  • A batch completes — a milestone’s worth (or a ready-backlog slice) of tasks executes as stacked or non-overlapping parallel PRs.

Deliverables

Section titled “Deliverables”
  • The executable-task lifecycle operations on the TS substrate (the cap-1..cap-10 surface).
  • The orchestrate dispatch loop and PR-state classification.
  • Git-ref lease coordination (per the github-ref-leases ADR).
  • Task kind + leaf-only dispatch wired into selection (D-VSLI-distributed-work-runner-architecture).

Out of scope

Section titled “Out of scope”
  • The product definition itself — M-K7DP-initial-product-definition.
  • Full multi-product orchestration and production hardening beyond a working prototype.
  • Deep milestone-batching strategy (stacked-PR rebase semantics, runtime file-overlap analysis) beyond what the prototype needs.

Risks / open questions

Section titled “Risks / open questions”
  • Depends on the TypeScript substrate landing; the prototype can only harden as the port completes.
  • “Prototype” scope is easy to inflate toward production — keep the bar at “the loop demonstrably runs,” not “the loop is bullet-proof.”

← Back to Milestones