Stop the dream cycle aborting when memory consolidation returns early

[rockfordlhotka]

Summary

The memory consolidation pass ran inline at the top of DreamAsync, and its two early exits — fewer than two stored entries, or a null result from the LLM — used a bare return. That returned from the whole dream cycle, not just the pass, so every pass behind it was silently skipped.

The worst case is a freshly provisioned agent. With an empty memory store, consolidation hits the all.Count < 2 guard on every cycle, so memory mining never runs and the store can never reach two entries. It is a deadlock: the agent cannot mine its first memory, and nothing in the logs says why — the cycle logs as started but never logs as complete.

// before — inside DreamAsync, so this returns from the cycle
var all = await _memory.SearchAsync(new MemorySearchCriteria(MaxResults: 1000));
if (all.Count < 2) { ...; return; }   // <- skips episode, entity, mining, everything

Extracted into RunMemoryConsolidationPassAsync, returning its deleted/saved counts, so the early exits return from the pass and the cycle continues.

Most of the diff is re-indentation from that extraction; reviewing with ?w=1 shows the real change is ~100 lines.

Also in this change

Two dream-subsystem fixes found while scoping the above. Calling them out rather than burying them:

• Dream:MemoryConsolidationEnabled (default true, so existing deployments are unaffected). Consolidation is the only pass that rewrites stored entries rather than only adding or deleting them, which makes it the only place a dream model can introduce detail that no source entry contained — and since it reads the memory store as its own input, anything it invents is re-ingested as fact next cycle. Mining, episode and entity extraction all derive from the conversation log and can be checked against it. Turning it off costs duplicate merging, importance re-scoring and decay, all of which live inside the pass.

• Dream:ModelTier (default Balanced, matching previous behaviour). One pass was hardcoded to ModelTier.Balanced while the rest of the cycle honoured the configured tier. Dream passes are structured extraction — read a transcript, return JSON — so pointing them at a different tier from the conversational one is useful independently of the bug.

Testing

• New DreamMemoryConsolidationToggleTests: default-enabled, environment-style binding (Dream__MemoryConsolidationEnabled), and independence from Dream:MemoryMiningEnabled.
• Full RockBot.Host.Tests suite passes (1080), verified with only this change applied and rebased onto current main.

Note on the first commit

This branch also carries a one-line .gitignore fix. deploy/docker-compose/agent-data/ is the bind-mounted runtime directory for the compose agent — profile markdown, long-term memory store, knowledge graph, conversation transcripts. It was untracked but not ignored, so git add -A would stage the entire local agent state. It has never been committed; this only prevents it. agent.extra.env is ignored alongside it for the same reason as the existing .env entry: it can carry provider credentials.

Happy to split that into its own PR if preferred — it is unrelated to the dream fix, but it seemed worth landing promptly.

🤖 Generated with Claude Code