Blog
Notes on planning documents, AI agents, and building this in the open.
guideplansclaude codecodex
Version control for Claude Code and Codex plans
Where Claude Code and Codex keep the plans they write, what is left of them a month later, and two ways to keep every version: git, or flanner.
guideplansclaude code
What Claude Code's plansDirectory setting does, and does not
plansDirectory moves Claude Code's plan files into your project. How to set it, what it changes, and the four jobs it still leaves to you.
guideplansfreshness
How to spot a stale plan before your agent follows it
The signs that a plan no longer matches the code, how to check each one with git, and how flanner freshness checks them for you and shows the evidence.
guideagentsclaude codecodex
How to keep CLAUDE.md and AGENTS.md in sync
Claude Code reads AGENTS.md only when no CLAUDE.md exists. The one-line import that keeps a single source, when a symlink is enough, and what flanner writes.
guideskillsclaude code
Claude Code skill not loading? Find the copy that wins
Why an edited Claude Code skill does nothing: where skills load from, which copy wins a name clash, the description budget, and how to find the copy in effect.
guidememoryclaude codemesh
How to sync Claude Code memory across machines
Claude Code's auto memory stays on the machine that wrote it. The workarounds people use to share it, what each costs, and how flanner shares chosen memories.
visionplansagents
Plans that stop being true
An agent writes a good plan in ninety seconds. Three weeks later a third of it is wrong and nobody knows. That is the problem flanner is for.