Back to Skills

wayfinder

把单个 agent session 装不下的一大块工作规划成 issue tracker 上的 decision tickets shared map,并逐一解决,直到通往 destination 的路清晰。

2,892stars236forksUpdated 8/5/2026

Security Assessment

Safe(94/100)
Security Score94/100

About wayfinder

wayfinder (documented in Chinese) is a planning skill for breaking a large piece of work that cannot fit in a single agent session into a shared "map" of decision tickets on an issue tracker, then resolving them one at a time until the path to a destination is clear. It solves a scoping-and-continuity problem: an idea that is too big and "fogged" to execute directly gets charted into decisions rather than build tasks, so the route is discovered incrementally instead of rushed. Its explicit default is "plan, don't do": each ticket resolves one decision, and the map is complete when nothing remains to decide before someone starts building.

The map is a single issue labeled wayfinder:map, acting as an index (not a store) with sections for Destination, Notes, Decisions-so-far, Not-yet-specified (fog), and Out-of-scope; tickets are its child issues, each a single question sized to fit within roughly a 100K-token session. Tickets carry a type label — research (AFK, delegated to a /research subagent), prototype (HITL), grilling (HITL, the default), or task (HITL or AFK) — and are claimed by assignment, blocked via the tracker's native dependency relationships, and resolved with answers recorded at resolution rather than pasted into the body. The skill defines a "fog of war" model distinguishing tickets you can specify now from fog you cannot yet, plus explicit out-of-scope handling, and two invocation modes: charting a new map and working through an existing one (never resolving more than one non-research ticket per session).

The target users are engineers and other planners tackling ambiguous, oversized efforts — engineering work, course content, or anything of that shape — who want a durable, collaborative planning structure on their issue tracker. It is a text-only methodology skill with no executable or risky components.

FAQ

What is the core idea of wayfinder?

To chart a large, fog-surrounded effort into a shared map of decision tickets on an issue tracker and resolve them one at a time, finding the path to a destination rather than rushing to build.

Does it execute work or just plan?

By default it only plans — each ticket resolves one decision and produces decisions, not deliverables. An effort can override this in the map's Notes to bring execution into scope.

What ticket types exist?

research (agent-driven, handled by a /research subagent), prototype (human-in-the-loop), grilling (the default human-in-the-loop conversation), and task (the only type that does work rather than deciding, to unblock a decision).

How is it invoked?

Two modes: 'chart the map' when a user brings a loose idea, and 'work through the map' with an existing map URL or number. In either mode a session resolves at most one non-research ticket.

What if there is no fog when charting?

If there is no fog, the path is already clear and the whole journey fits in one session, so you do not need a map — the skill says to stop and ask the user how to proceed.

All Files

2 files
agents/openai.yaml0.1 KB
View
SKILL.md7.2 KB
View

Install wayfinder

Download and extract the skill files to your .claude/skills/ directory.

Quick Setup:

  1. Copy the skill folder to .claude/skills/
  2. Claude will automatically detect and use the skill