New /wayfinder skill aims to reduce planning friction for autonomous agents

Matt Pocock, creator of the ‘AI Skills for Real Engineers’ project on GitHub, released a new skill called /wayfinder to address planning challenges in agentic workflows. The tool is designed to help agents navigate ambiguous project scopes, which Pocock describes as navigating ‘the fog of war’—a state where the end goal is unclear at the outset.

/wayfinder introduces a multi-session planning layer that splits planning into separate sessions: grilling (alignment), prototyping, and research. This structure aims to reduce constraints imposed by context windows and session management, allowing agents to handle more complex planning tasks without manual intervention.

The skill uses a structured terminology system to reduce ambiguity in agent communication. Pocock emphasizes the importance of ‘leading words’—precise terms like ‘map’, ‘ticket’, and ‘session’—to guide agents through the planning process. This approach is part of a broader effort to create a consistent ‘AI coding dictionary’ for agent interactions.

/wayfinder is positioned as a complement to Pocock’s existing ‘grill-me’ skill, which is intended for smaller, well-defined tasks where the entire plan can fit into a single session. In contrast, /wayfinder is designed for scenarios where the path forward is unclear, such as rearchitecting a personal website with decades of content or planning courses.

Pocock’s design process for /wayfinder involved experimenting with centralized documents to manage context and information flow. The tool’s architecture reflects this focus, with distinct roles for ‘map’ (overarching decisions), ‘ticket’ (specific tasks), and ‘session’ (execution units) to ensure clarity and reduce confusion in agent behavior.