Goal-scoping skill — convert a GOAL into owned LANES (the missing epic-lifecycle front-end)

Context

friction→gold for a session-long, repeated planning failure (operator-surfaced 2026-06-27). v13.1 had no defined goals + no owned lanes, so the team chased scattered micro-tickets, parents never closed, and a peer reported "nothing left to do." Two symmetric failures recurred all session: (a) no planning → micro-ticket chase; (b) lead micro-manages → assigns peers to lanes (violates self-assignment + flat-peer agency). The root: we have no planning skill. /epic-create exists but only authors one epic; nothing scopes a GOAL into the set of LANES.

The Problem

The epic lifecycle has three skills — epic-create (author one epic), epic-review (pre-work), epic-resolution (closeout) — but no front-end: nothing converts a release/initiative GOAL into the coherent owned LANES that become those epics. So scoping is done ad-hoc — either skipped (→ scrap-ticket sprawl) or done as lead-assignment (→ micro-management). Both are wrong.

The Architectural Reality

• Epic lifecycle family: .agents/skills/epic-create / epic-review / epic-resolution. Each operates on ONE already-identified epic.
• Missing: the goal → {lanes} scoping step that decides which epics/lanes a goal needs. This skill is that front-end; each lane it produces is authored via epic-create.
• Ownership model: peers self-select lanes (flat-peer agency, AGENTS.md §swarm_topology_anchor); the planner defines goal+lanes but never assigns a peer (the lead-as-lane-assigner anti-pattern).

The Fix

A new goal-scoping skill (Progressive-Disclosure: thin SKILL.md router + references/goal-scoping-workflow.md payload). The process:

1. Define the GOAL — one demonstrable bar.
2. Scope into LANES — a few (≈2–6) coherent owned streams (by subsystem/daemon/pillar/capability), each a whole unit worth a dedicated owner — not slivers, not scrap-tickets, not a flat ticket list.
3. Each lane → an epic via /epic-create — the durable, context-wipe-proof ownership anchor.
4. Ownership = self-select — peers claim lanes; the planner defines + facilitates, never assigns.
5. Drive to the lane-GOAL (close-by-goal), decomposing reasonably (no micro-slivers).

Acceptance Criteria

• goal-scoping skill scaffolded: SKILL.md (7–12 line router) + references/goal-scoping-workflow.md payload, per the authoring guide.
• Payload codifies the goal→lanes process + names the two symmetric anti-patterns (no-planning / lead-assigns-peers) + the self-select ownership rule.
• Documents the epic-lifecycle relationship (goal-scoping = the front-end → epic-create → epic-review → epic-resolution).
• .agents/skills/skills.manifest.json updated + .claude/skills/goal-scoping symlink created.
• node ai/scripts/lint/lint-skill-manifest.mjs --base origin/dev passes.

Out of Scope

• The review + ticket-scoping right-sizing (#14263 — sibling process-reform).
• Re-authoring epic-create/review/resolution (this is the front-end that feeds them; they're untouched).

Avoided Traps

• "Extend /epic-create" — REJECTED: epic-create authors ONE epic; this scopes a goal into the SET of lanes (a distinct front-end function).
• "The planner assigns peers to lanes" — REJECTED: that's micro-management + violates self-assignment; peers self-select.
• "Scope the goal into a ticket list" — REJECTED: lanes (coherent owned streams), not scrap-tickets.

Related

• #14263 (sibling: review + ticket-scoping right-sizing). Both are the process-substrate reform.
• Builds on the epic-lifecycle skills + AGENTS.md §swarm_topology_anchor (flat-peer self-select).

Origin Session ID

3f32bbc7-1bfe-4f85-9232-c957de0d22f1

Authored by Vega (Claude Opus 4.8, Claude Code) as v13.1 lead/planner.