- Rust 2024
- Cargo virtual workspace
- serde + serde_json
- thiserror
- uuid (v1, v4 + serde features)
- proptest
- tokio (for future async)
- GitHub Actions CI
- Apache 2.0 license
- SQLite (planned, Phase 2)
- PyO3 (planned, Phase 12)
- napi-rs (planned, Phase 13)
- axum (planned for proxy + dashboard)
- ~/projects/active/evolve/Cargo.toml: Workspace root, virtual workspace
- ~/projects/active/evolve/rust-toolchain.toml: Stable channel
- ~/projects/active/evolve/.gitignore
- ~/projects/active/evolve/LICENSE (Apache 2.0 with copyright filled)
- ~/projects/active/evolve/README.md
- ~/projects/active/evolve/CONTRIBUTING.md
- ~/projects/active/evolve/.github/workflows/ci.yml: Rust matrix + coverage
- ~/projects/active/evolve/crates/evolve-core/Cargo.toml
- ~/projects/active/evolve/crates/evolve-core/src/lib.rs
- ~/projects/active/evolve/crates/evolve-core/src/schema.rs (FieldSchema enum, GenomeSchema, validate)
- ~/projects/active/evolve/crates/evolve-core/src/ids.rs (ProjectId/ConfigId/ExperimentId/SessionId/SignalId/AdapterId newtypes)
- ~/projects/active/evolve/crates/evolve-core/src/agent_config.rs (AgentConfig, ModelPref, ResponseStyle, default_for, fingerprint, extension)
- Tool-hook driven architecture for passive drop-in evolution
- Universal AgentConfig with typed extensions and opaque per-tool JSON map
- Champion/challenger Bayesian promotion for selective evolution
- Pivoted from active-evolution-on-benchmark to passive drop-in evolution
- Used champion/challenger Bayesian beta-binomial posterior for promotion
- Adopted universal AgentConfig core with typed extensions and opaque per-tool JSON map
- Tool-hook driven, no background daemon for zero idle cost
- Local-only SQLite at ~/.evolve/evolve.db for simplicity
- Single LLM client crate used for occasional challenger generation
- Dropped HTTP/gRPC service mode crate
- Virtual workspace pattern chosen over single root crate
- Adopted Rust 2024 edition
- Initial attempt to add a root [package] block in Cargo.toml resulted in architectural risk for the planned 7-crate workspace
- Reverted to virtual-workspace shape with members=['crates/evolve-core'] after spec reviewer feedback
- Original active-evolution architecture was fully designed before product pivot revealed it was the wrong shape
- Initial subagent-driven execution loop was overkill for byte-exact scaffolding tasks; switched to direct execution for trivial scaffolding