turlockmike

Remove specific observations from long-term memory

Force extraction of observations from the current conversation into observational memory

Search and display observations from long-term observational memory

Compress and reorganize observational memory, condensing old observations

MVM offline integration. Reads mvm stats + recall-log; resolves every detected anomaly within the same cycle (no sinks). Coverage → proactive ingest. Quality failure → cross-check → re-ingest if confirmed. Staleness → cross-check → supersede if confirmed. Contested/gap → user escalation. Triggers: '/mvm-dream', 'mvm dream', 'mvm consolidate', 'mvm maintenance'.

MVM closed-loop write. Reads a source (URL/file), authors locked test cases, runs naked baseline + injected verify (Agent-tool cold-clones, model: haiku), commits canonical only if all tests pass. Triggers: '/mvm-ingest', 'ingest <source>', 'add to mvm KB', 'remember this from <url>', 'capture <source>'.

MVM unified question-answering. Step 0 memo-cache, then KB + web retrieval INLINE (orchestrator searches itself) plus ONE naked-baseline haiku probe for prior-contamination signal; reconciles via trust hierarchy KB>web>weights, auto-spawns /mvm-ingest on web hits, logs every recall via `recall-log add`. Triggers: '/mvm-recall', 'recall <question>', any factual question routed to mvm.

Use when building CLI tools that need to work with AI agents, adding --agent flags to tools, making tools agent-compatible, or implementing POSIX Agent Standard (PAS) compliance. Use when the user wants machine-readable output from CLI tools or asks about JSON output for automation. CRITICAL - Use when building scripts for skills - instead of multiple individual scripts, build one PAS-compliant CLI with subcommands for easier maintenance, testing, and composability.