Repo context

Veralith never sees your codebase — but during a heal, your agent is inside it. Repo context captures what the agent learns and feeds it back into future diagnoses, so recommendations get smarter over time without you doing anything extra.

How it works

While working a heal, the agent calls save_repo_context with three fields:

• summary — one line on what the app/repo is.
• architecture — the stack, RAG pipeline, data sources, key files, and constraints.
• recommendations — repo-specific guidance for future fixes.

Veralith stores one repo-context record per project and injects it into the suggester's prompt when it writes the next heal card.

Why it matters

A concrete example: if the agent notes "the corpus is a static snapshot, so live-data queries are genuinely out of scope", then the next time a live-data question fails, Veralith's recommendation is "name the gap deterministically / abstain" rather than "improve retrieval" — because it now knows retrieval can't fix an out-of-scope question. The insight the agent gave once becomes an automatic part of every future diagnosis for that project.

You don't do anything

Repo context is captured by the agent as part of the heal flow — there's no separate step for you. Each heal refreshes it, so it stays current as your codebase evolves.