This is a practical index for "where should I look first?".
For step-by-step execution flow, read runtime-walkthrough.md first.
orchestrator.rs: recursion, execution strategy, synthesis controlcomplexity.rs: activation heuristics and mode selectionrepl.rs: sandboxed execution orchestration
context/: session state, externalization, typed prompt plumbingsignature/: typed signatures, validation, fallback extraction, submit semantics
llm/client.rs: provider callsllm/router.rs: model selection/routingllm/batch.rs: batched query executionllm/cache.rs: cache behavior
memory/: hypergraph persistencereasoning/trace.rs: decision graph representationreasoning/query.rs: trace analysis and traversal
spec_agent/: NL intake/refine/formalize/verify workflowsync/: Topos/Lean drift detection and dual-track operations
epistemic/: claim verification/confidence analysisproof/: Lean-oriented proof workflows and tacticsdp_integration/: quality review/coverage/proof status tooling
adapters/: external integrations (Claude Code, TUI, CLI)ffi/,pybind/,go/: language interoperability surfaces
- Activation behavior wrong?
- Start at
complexity.rs, thenorchestrator.rs.
- Structured output broken?
- Start at
signature/validation.rs,signature/submit.rs,signature/fallback.rs.
- Drift reports odd?
- Start at
sync/drift.rs, thensync/types.rs.
- Memory inconsistency?
- Start at
memory/store.rs, thenreasoning/store.rs.
- Keep changes scoped to one concern where possible.
- Add tests in same area as changed behavior.
- Update docs for behavior changes in same commit.
- Preserve integration contracts unless explicitly changing them.
A map is not the territory, but it does reduce wandering with a flashlight.