Agent4Agents (Project Compact) tackles four common failure modes in agentic coding — bogus work, architectural re-litigation, decision fatigue, and token waste — by enforcing project rules mechanically rather than relying on prompts. It uses a cache-aligned context prefix, append-only architectural decision records (ADRs), tier-adaptive ceremony, and server-side enforcement via compact-check in CI with no admin bypass. Written in Python; pairs with gstack as the workflow engine. 