/.github/loop/PRIORITIES.md

Priorities

The ranked work queue for the autonomous improvement loop. The architecture-review pass (the architect) owns this file: each run it turns the roadmap plus an internal scan (gaps in the services → agents → workflows lifecycle, API coherence, drift, tech debt, test and DX friction) into a single ordered list — highest-value first — and links each item to a tracking issue. The hourly continuous-improvement pass works the top item whose issue is still open. So the architect decides what, and the increment loop builds it.

Reading / editing. An item is done when its linked issue closes (the increment that builds it adds Closes #<issue>). Roadmap phase (Now → Next → Later) is the primary ordering; internal findings are interleaved by value, not kept in a separate list. The human can reorder this list — or the issues — at any time to redirect the loop; direction always wins.

Off-limits to the loop (the architect proposes these as notes, never as queue items the loop can auto-merge): brand/positioning copy, breaking public-API changes, architectural rewrites. Those go to the human.

Work queue (ranked)

1. Make AtlasCloud conformance deterministically request the echo tool (#3924) — #3926 shipped the previous plan/delegate notify wait fix, so the queue should drop the completed #3918 item. The freshest live harness signal now fails even earlier: AtlasCloud/minimax-m3 completes without invoking the required conformance_echo tool. This is the highest Now-phase item because the getting-started and 0→hero contract depends on providers actually using service tools, not just returning prose, and the CI gate needs an actionable deterministic failure path.
2. Make AtlasCloud conformance deterministically exercise guarded delegation (#3917) — after the echo-tool path is stable, the same live provider must attempt the blocked delegate path so guardrails are proven under a real model, not only mocks. This keeps plan/delegate safety provider-portable without public API changes and protects the services → agents → workflows story from drifting at the delegation seam.
3. Broaden provider streaming and keep chat/A2A streaming end to end (#3903) — once the live gate is stable, streaming is the highest developer-visible Next-phase seam. Real chat and long-running A2A tasks need token streaming to stay coherent from provider → ai.Stream → micro chat → A2A message/stream, with mock/default CI coverage plus key-gated live provider checks and safe fallback for non-streaming providers.
4. Trace agent runs as OpenTelemetry spans (#3908) — the blog/README/roadmap story promises an operable harness, and the developer on-ramp now includes chat, inspect, and run-history checkpoints. The next observability gap is production-grade trace correlation for RunInfo: steps, tool calls, delegation, status, durations, and failures should be visible as spans while defaulting to no-op when tracing is not configured.

Seeded by Claude Code from the roadmap + open issues; thereafter maintained by the architecture-review pass.