Problem
tests/ contains a single test file. There is no coverage of the agents, the routing logic, the governance layer, or the memory subsystem.
Proposed solution
Prioritise by risk:
governance/ — policy enforcement and cost management. Policy bugs are the expensive kind; test the deny paths specifically.orchestration/ — routing decisions. Given this state, does the graph go where it should?agents/ — one behavioural test per agent with the LLM mocked.memory/ — retrieval relevance and persistence round-trips.
Add per-agent evals with fixed inputs and expected behaviour, so LLM changes surface as measurable regressions instead of vibes.
Problem
tests/contains a single test file. There is no coverage of the agents, the routing logic, the governance layer, or the memory subsystem.Proposed solution
Prioritise by risk:
governance/— policy enforcement and cost management. Policy bugs are the expensive kind; test the deny paths specifically.orchestration/— routing decisions. Given this state, does the graph go where it should?agents/— one behavioural test per agent with the LLM mocked.memory/— retrieval relevance and persistence round-trips.Add per-agent evals with fixed inputs and expected behaviour, so LLM changes surface as measurable regressions instead of vibes.