/.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. CI-verify first-agent docs wayfinding links (#4291) — Developer adoption is the current goal, and the README/website/examples now tell a coherent first-agent path: install troubleshooting → micro agent demo/micro examples/micro zero-to-hero → examples index → no-secret first agent → debugging → 0→hero. The highest-value next guard is preventing that on-ramp from drifting. Keep this scoped to a local-file check wired into an existing harness or Make target, with no new runtime dependencies.
2. Fix AtlasCloud agent conformance guarded delegation (#4296) — The plan-delegate notify recovery shipped in #4299, but the latest live provider conformance run exposed a separate AtlasCloud agent harness failure: missing guarded delegate behavior within the retry budget. This is the top remaining Now-phase reliability defect because guarded delegation is core to services → agents handoff safety and should be CI-verifiable without public API changes.
3. Fix AtlasCloud A2A streaming fallback tool-call reporting (#4297) — The same live provider run found the A2A streaming fallback path reporting neither tool-call evidence nor run metadata (tool=false runInfo=false). Rank it just behind guarded delegation: A2A streaming is an important interop seam, but the issue is narrower than the base agent conformance failure and should stay focused on surfacing tool/run evidence in the existing harness.
4. Propagate cancellation and retry signals through provider model calls (#4273) — After the adoption link guard and current live AtlasCloud failures are stable, the remaining Now-phase resilience seam is provider-boundary cancellation/deadline/retry behavior across the agent loop. Keep it focused on fakes/tests and avoid public API or architecture changes.

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