GuardianKane: It Will Not Let Your Agent Lie
Two parts on the full build: the twelve-phase rebuild, and the visual layer that makes verification something you can see.
After the hackathon, GuardianKane went from a Stop hook to a reviewed, phased loop with a scope guard, a file lock, a secret scan, and a dashboard of code, memory, and PRD graphs with a chat panel wired into the live Claude Code session. Part 1 is how the rebuild happened and why most of it was wiring. Part 2 is the tour, including the bugs the dashboard exposed in the gate itself.
Start with part 01 →
- 01
Sep 21, 2026 · 8 min readThe Second Build Was Mostly Wiring
GuardianKane started as a PRD-to-task tool, became a Stop-hook gate, and then got rebuilt across twelve numbered phases. The lesson of the rebuild: most of it was connecting data that Kane CLI or my own hooks already produced, and the first phase fixed a bug that broke every fresh install.
- 02
Sep 21, 2026 · 11 min readMaking Verification Visible
A pass or fail line at the end of a turn is not enough to trust an agent's build. A tour of GuardianKane's dashboard: the code, memory, and PRD graphs, the Kane activity feed, a chat panel wired into the live Claude Code session through an undocumented socket, and the bugs the dashboard exposed in the gate itself.