/.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. Prevent checkpointed tool calls from panicking (#4594) — Highest-value Now-phase resilience gap: a live AtlasCloud plan/delegate run completed the intended side effects but then crashed inside checkpointed tool handling. The services → agents → workflows lifecycle cannot feel operable if malformed or repaired provider output can SIGSEGV the harness instead of returning a classified error, retrying, or failing safely.
2. Thread first-agent quickcheck through docs wayfinding (#4599) — Keep developer adoption weighted alongside hardening. The CLI quickcheck breadcrumbs shipped in #4597, but the README/website first-agent path should make the recovery command discoverable before a new user falls into the full docs tree. This preserves the scaffold → run → chat → inspect on-ramp as a contract, not tribal CLI knowledge.
3. Handle incomplete AtlasCloud agent-flow workspace tool calls (#4595) — Next resilience gap from the same live provider-conformance run: the event-driven agent-flow path achieved the expected workspace/notification side effects, then still surfaced a failed flow on an incomplete repaired text tool call. Fix after the panic guard so successful/suppressed side effects do not leave the workflow story looking failed.
4. Make AtlasCloud universe A2A reachability probe deterministic (#4504) — The remaining live AtlasCloud interop seam is the universe A2A reachability probe intermittently timing out. Keep it in the queue, but behind the crash/tool-call correctness issues and the adoption quickcheck follow-through, because false-negative reachability matters most once the core live flow exits cleanly.

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