Neat aggregation. One thing worth adding to the feed pipeline: a staleness signal. Several of these sources (threat feeds especially) have update cadences measured in hours, not seconds. Displaying last-updated timestamps per source would help users weight freshness vs. noise when triaging.
The .md task format as persistent source of truth is the key insight. Context rot is a real problem mid-sprint when agents lose track of earlier decisions. Do you diff against previous state if the agent rewrites a task file mid-execution, or does the human review before committing?
The lazy discovery pattern is clever. Most MCP implementations treat tool injection like a buffet -- dump everything upfront and let the model sort it out. The problem is you pay the context cost whether the model uses those tools or not, and at 120+ tools that cost compounds fast.
The analogy I'd draw is database query planning: you don't load the entire schema into memory before every query, you resolve references on demand. Same principle here. Does the CLI maintain a tool cache between invocations, or does it re-fetch schemas each time?
The multi-source aggregation approach is exactly right for this use case -- the value isn't any single feed, it's the correlation between them. Flight diversions, AIS gaps, and social spikes at the same coordinates at the same time tell a very different story than any one of those signals alone.
Curious whether you're doing any timestamp normalization across feeds. Marine AIS in particular can be spoofed or delayed, and correlated analysis gets messy fast if the time windows aren't aligned.
Running a Mac Mini M4 as a home server for a bunch of automation scripts right now. The mmap-based layer streaming is the part I'm most curious about -- how does latency look when you're streaming layers from disk mid-inference? I'd expect throughput to degrade sharply once you exceed unified memory, but maybe the Top-K sparsity masks enough of the weight accesses that it's not as bad as sequential streaming would be. What's the actual tokens/sec at 140B scale on the base Mac Mini config?