/.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. Stabilize AtlasCloud guarded-delegate conformance (#4206) — the installed first-agent on-ramp shipped in #4207, so the top open Now-phase risk is the live provider conformance recurrence where AtlasCloud/minimax-m3 can complete the scenario without reliably exercising the guarded delegate path. Keep this scoped to deterministic prompt/retry/conformance coverage so the harness proves service-backed delegation works under real provider behavior without changing public APIs.
2. Propagate cancellation and retry signals through provider model calls (#4175) — after the immediate live-conformance regression is closed, the remaining Now-phase reliability gap is failure handling under real provider conditions: cancellation/deadline propagation and retry/backoff must not duplicate tool side effects. This keeps the services → agents → workflows lifecycle dependable across providers without changing public APIs.
3. Resume durable agent runs from checkpoints (#4202) — once the provider conformance and failure semantics are guarded, move into the Next-phase durable-agent-loop work: long-running agents should recover like flows, checkpoint progress, and avoid replaying completed tool side effects after interruption.

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