A Line of Documentation Was Acting as a Global Config Flag
I spent a morning hunting for a setting that did not exist. A while back I turned off Claude's co-authorship trailer in my commits — a deliberate choice at the time. Last week I decided I wanted it back. So I went looking for the switch I'd flipped. ~/.claude/settings.json : no key. settings.local.json : no key. ~/.claude.json , parsed as JSON, top-level plus all 38 project entries : no key. The…
A sentence in a rules file was acting as a global configuration flag in a personal multi-agent system built on top of Claude Code. Despite the claim that a global setting was off, the prose was the switch controlling the model's behavior. The feature was on, but the description of the world said otherwise. This highlights how in an agent system, prose and constants are not just documentation about the control plane; they are the control plane themselves.
Rules files get loaded into every session, making any claim inside them have the reach of a global setting. The model doesn't verify these claims against the system, as from inside the loop, the claim is the system. A frozen figure in a document generation script was incorrectly set to 617K, despite the correct figure being 577K.
The issue originated from a literal in the system prompt, and even though the correct figure was updated, the stale value remained in a generated artifact downstream. This demonstrates how a prose rule that holds "I want to be fair" can lead to errors if the source of truth is not properly updated.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
