wayfind

Pre-computed context

Current user: !gh api user --jq '.login' 2>/dev/null || echo "unknown" Open maps: !gh issue list --label "$(l=$(jq -r '.config.container_label | if type=="string" then . else "" end' .work-item-tracker.json 2>/dev/null); echo "${l:-work-map}")" --state open --json number,title --jq '.[] | "#\(.number) \(.title)"' 2>/dev/null || echo "none"

Variables

Arguments: $ARGUMENTS

Purpose

Some efforts are too big to hold at once AND too foggy to ticket. You can't yet phrase half the questions, let alone answer them. /planning:interview needs a coherent task; /planning:plan needs a coherent plan; both presuppose you already know what you're deciding. /planning:wayfind sits upstream of all of them: it turns a too-big-foggy effort into a shared decision map on the work-item tracker, then works that map's frontier one decision at a time until the fog burns off and a real destination (Brief / PRD / PLAN) can be handed onward.

Plan, don't do. A map holds decisions, not build work. Each decision item, once resolved, either sharpens the map or graduates to the destination. The moment the destination is coherent, the map closes and the normal pipeline (/planning:interview → /planning:design → /planning:plan → /implementation:implement) takes over. The map persists as native tracker primitives, each decision routes to a first-party skill, and execution artifacts live in <memory_dir>/<slug>/ (default .work/). The topic-docs convention's memory tier, slug spec and all (see ${CLAUDE_PLUGIN_ROOT}/reference/topic-docs.md) never in the map itself.

Two modes. chart builds or extends a map (interactive only). work picks one item off the map's frontier and drives it to resolution. The default action auto-detects: an existing open map for the topic → work; nothing yet → chart.

The fog test (the one call that governs everything)

For every uncertainty, ask: can I phrase it as a sharp question?