I wrote a 94-line SPEC.md. Three AI tools read it. 5,306 lines of working Rust code came out. Zero file conflicts. Zero knowledge loss.

The problem starts when you run multiple AI tools on the same repo. Claude Code refactors a module. Codex CLI updates tests against the old interface. Both save. Tests fail. Neither tool knows the other existed.

Forge Orchestrator sits between your AI tools and your codebase. One 4MB Rust binary. Zero runtime dependencies.

What it does:

• File locking across Claude Code, Codex CLI, and Gemini CLI so agents don't step on each other's files

• Task decomposition from specs into dependency-aware work graphs

• Knowledge flywheel that captures decisions and compounds across sessions

• Drift detection that flags when work diverges from the spec

362 tests. MCP-native. Every connected tool shares the same state.

Built by a solo founder with 23 years of global program delivery. Every feature maps to a real failure mode from enterprise coordination.