feat: add chat activity status

[casey-brooks]

Summary

• add runners/identity integration for GetChats activity_status, active_workload_ids, and unread_count
• update MarkAsRead to ack all unacked messages via Threads paging
• extend config/proto generation to include runners/runner/identity

Testing

• go vet ./...
• CHAT_E2E_INPROCESS=1 go test ./...

Issue

• Relates to Implement chat activity status + per-chat unread counts + workload room subscriptions architecture#142

[casey-brooks]

Summary

• add runners/identity integration for GetChats activity_status, active_workload_ids, and unread_count
• update MarkAsRead to ack all unacked messages via Threads paging
• extend config/proto generation to include runners/runner/identity

Tests

• go vet ./...
• CHAT_E2E_INPROCESS=1 go test ./...

Results

• Tests: passed (failed: 0, skipped: 0)
• Lint: no issues

[noa-lucent]

Static review only (did not run tests locally).

Requesting changes:

• GetChats should make exactly one Threads.GetUnackedMessageCounts call per request/page (architecture#142). Today it’s inside the ListChats loop, so it can be called multiple times when filling pages after filtering.
• GetChats should fan out to Runners.ListWorkloadsByThread in parallel per (thread, non-passive agent) pair (with a concurrency limit). Current implementation is sequential.

Minor/nits:

• Consider adding a unit test assertion for the single GetUnackedMessageCounts call (the fill-after-filtering test is a good spot).
• E2E test name TestMarkAsRead_EmptyMessageIDs no longer matches behavior.
• PR description: please avoid ambiguous “#142” references; prefer explicit agynio/architecture#142.

[casey-brooks]

Summary

• call Threads.GetUnackedMessageCounts once per GetChats request and assert in fill-page test
• parallelize Runners.ListWorkloadsByThread fan-out with concurrency limit and aggregate results per thread
• rename e2e test for MarkAsRead no-unacked case

Tests

• go vet ./...
• CHAT_E2E_INPROCESS=1 go test ./...

Results

• Tests: passed (failed: 0, skipped: 0)
• Lint: no issues

@noa-lucent please re-review.

[noa-lucent]

Static review only (did not run tests locally).

Thanks — the original blockers look addressed:

• GetUnackedMessageCounts is now called once per GetChats request, and there’s a unit test assertion for it.
• Runners fanout is now parallelized with a concurrency limit.
• E2E test was renamed to match the new MarkAsRead contract.

Remaining blockers:

• Runners partial-failure handling regressed: if any (thread,agent) workload lookup fails, the whole thread becomes CHAT_ACTIVITY_STATUS_UNSPECIFIED. Per agynio/architecture (chat.md partial failure handling + #142), failures should degrade that pair to finished (no workload IDs) while still aggregating other agents; if all pairs fail the chat should degrade to FINISHED (not UNSPECIFIED).
• workloadSummaryForAgent paginates through all workloads, which can produce many ListWorkloadsByThread RPCs per (thread,agent) pair for long-lived threads. This conflicts with the #142 acceptance signal (“one ListWorkloadsByThread call per pair on the page”). Prefer fetching only the most recent workload (ideally page_size=1) and avoid pagination.

[casey-brooks]

Summary

• treat per-(thread,agent) workload errors as finished while still aggregating other agents
• fetch only the most recent workload per agent with page_size=1 (single RPC per pair)
• assert ListWorkloadsByThread request params and add coverage for failure->finished behavior

Tests

• go vet ./...
• CHAT_E2E_INPROCESS=1 go test ./...

Results

• Tests: passed (failed: 0, skipped: 0)
• Lint: no issues

@noa-lucent please re-review.

[noa-lucent]

Static review only (did not run tests locally).

Re-review: fixes look good.

• Runners partial failures now degrade the affected (thread,agent) pair to FINISHED (without blanking the whole thread), and if all pairs fail the chat resolves to FINISHED with empty active_workload_ids.
• ListWorkloadsByThread is now a single call per (thread,agent) pair (page_size=1, no pagination), and unit tests assert this.

All my prior threads are resolved.

[casey-brooks]

Summary

• regenerated protobufs and committed generated identity/runners/runner Go outputs

Tests

• go vet ./...
• CHAT_E2E_INPROCESS=1 go test ./...

Results

• Tests: passed (failed: 0, skipped: 0)
• Lint: no issues

CI is running now; I’ll follow up once it’s green.

[casey-brooks]

CI update: build run https://github.com/agynio/chat/actions/runs/24945414052 completed successfully.

[noa-lucent]

Static review only (did not run tests locally).

Re-review after CI/proto generation fix: new commits only add missing generated proto Go files for identity/runner/runners. No behavioral changes to the server logic; prior contract/architecture concerns remain satisfied.