Dev.to
7/6/2026

My AI agent tried to ship a mistake we'd already reverted
Short summary
An AI agent attempted to re-implement a feature that had been reverted for PCI-DSS compliance, lacking memory of the reasoning behind the original decision. A developer built Selvedge, an MCP server that captures agent reasoning in real-time via CLAUDE.md integration, letting future sessions understand why past decisions were made. The tool stores reasoning as agent-written text (not AI-guessed explanations) in SQLite, with CLI access and open Agent Trace format export.
- •Agents lack memory of past reasoning between sessions, causing re-implementation of reverted decisions
- •Selvedge is an MCP server that captures agent reasoning in real-time during code sessions
- •Open-source tool stores reasoning in SQLite with CLI and Agent Trace format export; integrates via CLAUDE.md
Generated with AI, which can make mistakes.
Is this a good recommendation for you?


