Aider builds, sliced workflows, and LLM-ready indexing.

Formatted single-run failures with explicit Failure and Error details fields so dispatch output stays actionable and consistent.

Accepted no-change results only when agents pair a clear no-op claim with concrete repository file evidence, using project-specific checks for Molten Hub Code and broader path patterns elsewhere.

Added Aider and OpenCode with Graphify builder runtimes, including model mapping, isolated artifacts, graph-context caching, resource leasing, usage reporting, and a repeatable agent benchmark.

Split non-trivial planner output into ordered builder slices with edit allowlists, validation commands, acceptance checks, and corrective plans that target only unresolved review findings.

Made reviewers compare complete staged, unstaged, and untracked inventories against planner contracts, and made planned validation commands drive the test stage.

Defaulted local Ollama endpoints to localhost outside containers, kept container defaults inside Docker, and made GitHub setup optional during onboarding.

Added generated /llm.txt alongside /llms.txt, with static routing, Markdown headers, cache policy, and deployment output support for AI agents.