Claude-Mind: Skill for Claude Code

/sprint-planning — Fleet Sprint Planning Exercise

Turns the two report skills into a full planning pass: measure → analyze → verify → apply → publish. A plan that changes no queue state is a report, not a plan (asp-353 directive lineage) — but every queue change must survive a live-state verification first. Formalized 2026-08-10 from a user-directed sprint session (74-agent ultracode pass, 276 proposed changes, 28 refuted by adversarial verification — the refutation rate is why Phase 4's verify step is not optional).

Hybrid skill: user-invocable AND agent-callable (the recurring sprint-planning goal invokes it in standard mode). Requires assistant or autonomous mode — it writes queue state.

Sub-commands

/sprint-planning            — Standard pass: inline analysis, bounded lanes
/sprint-planning --ultra    — USER-INVOKED ONLY: authorizes a multi-agent
                              Workflow fan-out (per-aspiration analysts +
                              adversarial verifiers). An agent-initiated
                              (recurring-goal) run MUST NOT pass --ultra:
                              Workflow orchestration requires explicit user
                              opt-in, and a recurring firing is not one.

Phase 0: Load Conventions

Step 0: Load Conventions — Bash: load-conventions.sh with each name from the conventions: front matter. Read only the paths returned (files not yet in context). If output is empty, all conventions already loaded — proceed to next step.

Phase 1: Foundation Reports

1. Invoke Skill(backlog-report) — produces agents/<agent>/BACKLOG.md and the
   structural indexes this skill reuses (score map, blocked map, user goals,
   recurring health, testable hypotheses).

2. Priority dashboard data (inline — do NOT invoke /priority-review here; its
   Phases 3-4 are interactive and this skill's reorder decisions come from
   Phase 3 analysis instead):
   Bash: load-aspirations-compact.sh → Read returned path
   Bash: goal-selector.sh select → scored_goals (NOTE: output is a bare JSON
     array, not a dict) → aggregate score per aspiration
   Bash: echo '<[{asp_id,priority,score}...]>' | priority-review-mismatch.sh
     → flagged score-priority mismatches (needs 3+ consecutive runs to flag)

3. Snapshot for the ledger: record counts (active aspirations, non-terminal
   goals, selectable goals, blocked, user-routed, overdue recurring) BEFORE
   any write — these are the plan's before/after evidence.