The problem
Small AI teams increasingly use coding agents (Claude Code, Cursor, Copilot). Agents work best when internal docs are structured, machine-readable, and include examples. Most teams' internal docs were written for humans and fail when agents read them. The gap is painful but no standard pattern exists for 'agent-ready docs.' Build a reference implementation: take 3 open-source projects with existing human docs, produce an agent-ready version of each, build an eval harness that measures agent task success rate on the original vs agent-ready version. Open-source the pattern as a style guide and a conversion tool.
What success looks like
Agent task success rate 25%+ higher on agent-ready docs across 3 test projects. Published style guide + open-source conversion tool.
Data access
Public open-source projects (chosen with advisor input). TwoSetAI will provide internal ops docs as optional second test case.
Constraints
Open-source license required.
Reward
Featured in a TwoSetAI workshop + conference submission support