Issues

Browse common AI-era failure modes by category.

Agents

Agent actions, tool choices, and multi-step behavior.

Breakage caused by mismatched versions or stale expectations.

Missing instructions, examples, criteria, or fallback paths.

Repeated rules, overloaded fields, or steps doing too much.

Similar cases producing different results.

Work that cycles, waits, retries, or never settles.

Tool calls, integration boundaries, and downstream systems.

Stale, missing, restored, or leaking context.

Generated results that are malformed, unusable, or hard to verify.

Authority, ownership, approval, and action-control issues.

Conflicting rules, hidden instructions, or policy drift.

Work crossing task, file, context, permission, or action limits.

Small issues growing into wider downstream consequences.

Breakage between steps, owners, reviewers, or systems.