finding-your-unknowns
Finding Your Unknowns
When you work with an agent, the map is what you hand it (prompts, skills, context) and the territory is where the work actually happens (the codebase, real constraints). The gap between them is your unknowns — every time the agent hits one, it can only make a decision based on its best guess of what you want. The quality of the work is often bottlenecked not by the model's ability, but by your ability to clarify its unknowns.
This skill turns "find and clarify your unknowns" into an executable pre / during / post workflow, with a copy-paste prompt for every step. Core constraint: don't let the agent run a long task blind — spending the cheap effort to surface unknowns first is far cheaper than reworking later.
Method from Thariq Shihipar's (Anthropic) field guide Finding your unknowns (claude.com/blog, 2026-07-06). This skill is an executable packaging of it.
When to use
- You're starting a non-trivial agentic coding task (multi-file / unfamiliar codebase / unfamiliar domain / design work)
- A long-horizon task keeps coming back wrong — usually not the model failing, but unknowns left undefined
- You can't quite articulate what you want, or don't know what "good" looks like
- You're about to hand off a large agent-authored change for buy-in, or want to confirm you actually understand what it did