Docs Hill-Climb for Agents

Test and improve your docs with executable agent tasks so coding agents can follow them

Problem

More teams expect coding agents to follow their docs to do setup, CI checks and similar tasks, but docs are written for humans: missing steps, paths and preconditions. Agents fail repeatedly, and nobody has time to regression-test whether the docs actually work.

Solution

A local CLI that binds a set of tasks to your docs (e.g. "set up CI checks", "start the service locally"). Each run lets the agent attempt them and maps failures back to specific doc passages, generating rewrite suggestions and iterating in a hill-climbing loop. Outputs failure reports and doc diffs.

Users

Engineering leads rolling out coding agents, DevRel maintaining developer docs, indie builders working on agent toolchains

MVP

Task definition files; local execution with recorded agent failure steps; map failures to doc sections; generate rewrite suggestions and diffs; repeated runs forming a hill-climb history

Revenue

Open-source CLI plus paid cloud regression dashboard (per-project subscription) or per-run billing

Why now

Coding agents are becoming standard but doc quality is the bottleneck; open-source attempts validate the direction, and local execution with failure mapping is a differentiated, solo-buildable angle