retrospective

To document incident-driven learning, you capture user dissatisfaction details, analyze recent work against existing guidelines, and update AGENTS.md or development-principles.md to enforce explicit guardrails and prevent recurrence.

Updating development guidelines based on feedback requires analyzing misalignment against current principles, performing root-cause analysis on the failure, and codifying the resulting lesson directly into AGENTS.md to ensure future interactions meet expectations.

You need to perform a retrospective on project outputs when user expectations fail, decisions misalign, or post-release feedback indicates a gap in existing development principles that requires immediate process improvement.

Yes, this approach works for post-release feedback by logging the dissatisfaction, identifying missing principles that caused the failure, and applying explicit guideline changes to development-principles.md to prevent similar incidents.

To start a retrospective after an output fails expectations, describe the specific dissatisfaction, and the system will guide you through collecting details, analyzing root causes, and updating your guidelines accordingly.

A limitation when updating AGENTS.md with incident analysis is that it requires explicit user feedback detailing what went wrong; without clear dissatisfaction or context about the failure, the root-cause analysis cannot generate meaningful guideline changes.