Planning large-scale projects with AI agents often runs into context window limitations, forcing developers to downsize their ambitions to fit single-session constraints. Wayfinder is an open-source skill designed to solve this issue by orchestrating ambitious, multi-session planning through an interactive map of decision tickets in your existing issue tracker.

How Wayfinder Works: Mapping the Fog of War

Wayfinder takes a high-level destination (such as a feature specification or project goal) and breaks it down into structured decision tickets. It tracks dependencies and manages the “fog of war”—unresolved decisions that require research, conversation, or prototyping before moving forward. As each ticket is resolved in its own isolated agent session, the findings roll back up into the main map until the destination is fully realized.

The Four Core Ticket Types

To keep planning flexible and dynamic rather than falling into rigid waterfall patterns, Wayfinder utilizes four distinct ticket types:

• Grilling: Interactive Q&A sessions where the AI interviews you to clarify project requirements and design choices.
• Research: Automated information gathering executed in sub-agent sessions that report back with findings.
• Prototype: Quick, high-fidelity mini-builds used to test assumptions early and get practical feedback.
• Task: Real-world actions or manual execution steps required before proceeding.

Workflow and Execution

Once all decision tickets are resolved and the fog of war clears, Wayfinder compiles the decisions into a dense specification document linking directly back to original decision tickets as primary sources. From there, standard implementation tickets are generated and executed. Unlike traditional Spec-Driven Development (SDD), these generated specifications are non-persistent—once the feature is implemented into the code, the spec issue is closed and discarded.

Key Takeaways and Non-Coding Applications

Wayfinder is issue-tracker agnostic (working seamlessly with GitHub, Linear, Jira, or custom tools) and context-aware. Beyond software engineering, this dynamic multi-session planning methodology can be applied to complex real-world projects like course creation or offline project management.

Mentoring question

How do you currently handle project planning when a task is too complex or context-heavy for a single AI agent session, and how might dynamic ticket mapping change your context management strategy?

Source: https://youtube.com/watch?v=F3lL98Pj90o&is=9eDBwFqiE7mO-EPJ