16 KiB
id, type, title, domain, repo, status, owner, topic_slug, planning_priority, planning_order, created, updated, state_hub_workstream_id
| id | type | title | domain | repo | status | owner | topic_slug | planning_priority | planning_order | created | updated | state_hub_workstream_id |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CUST-WP-0046 | workplan | Activity-Core Hourly RecentlyOnScope Reports | custodian | the-custodian | finished | codex | custodian | high | 46 | 2026-05-22 | 2026-06-04 | 671153ff-55bc-4ace-aa97-f322ca76ab3c |
CUST-WP-0046 - Activity-Core Hourly RecentlyOnScope Reports
Goal
Move routine reporting onto owned activity-core infrastructure by scheduling an hourly RecentlyOnScope generation run for every State Hub domain that was active in the last hour.
The immediate operational outcome is:
- activity-core owns the hourly schedule and run telemetry
- State Hub owns domain activity selection and RecentlyOnScope report generation
- generated reports use the existing State Hub RecentlyOnScope API and report directory
- Codex app automation fallback is retired after the activity-core path is verified
This work is the practical bridge from "Codex app fallback" to "owned activity-core operating habit" without introducing another cron mechanism.
Current Automation State - 2026-05-22
Current verified state:
- Codex app automation
daily-state-hub-wsjf-triageexists and isACTIVE, scheduled daily at 07:20 Europe/Berlin. - Custodian ActivityDefinition
activity-definitions/daily-statehub-wsjf-triage.mdis stillenabled: false. - The local machine currently has only State Hub Postgres running in Docker; the local activity-core dev stack is not running.
- State Hub progress reports that activity-core was deployed to the Railiance01 K3s environment on 2026-05-22, with API health, worker schedule sync, Temporal, NATS, and tests verified.
- State Hub already has deterministic RecentlyOnScope support from
STATE-WP-0044:POST /domains/{slug}/recently-on-scope/GET /domains/{slug}/recently-on-scope/GET /domains/{slug}/recently-on-scope/{report_id}- default report root
reports/recently-on-scope/ - default report range
1h
- One current report artifact exists for
custodian:reports/recently-on-scope/custodian/20260522T081157Z--1h.md.
Scope
In scope:
- Add a State Hub batch/selection surface that can determine which active domains had activity in a given window and generate RecentlyOnScope reports for those domains.
- Add an activity-core ActivityDefinition owned by this repo for an hourly RecentlyOnScope run.
- Keep the scheduled run deterministic; no LLM is needed for the first version.
- Record activity-core run evidence, State Hub progress, and generated report metadata for every hourly run.
- Define retention/idempotency for hourly report artifacts.
- Retire or pause the Codex app automation fallback after the activity-core routine is verified and the daily triage cutover decision is explicit.
Out of scope:
- Replacing the existing RecentlyOnScope report template.
- Rebuilding RecentlyOnScope generation inside activity-core.
- Generating reports for inactive domains unless the operator explicitly asks for a full-domain sweep.
- LLM summarization of RecentlyOnScope reports.
- Automatically editing workplans, canon, or domain goals from the reports.
Design Direction
Preferred path:
- State Hub exposes a deterministic batch endpoint for RecentlyOnScope hourly runs. It owns active-domain detection because it already owns domains, topics, workstreams, tasks, decisions, progress events, and report storage.
- activity-core schedules an hourly ActivityDefinition that calls the State Hub batch endpoint. activity-core should not duplicate domain-selection SQL.
- The hourly run is idempotent by
(window, domain_slug, range), reusing the existing report id format when possible. - The batch endpoint returns report metadata and source counts. activity-core stores that as run audit context and posts a compact progress event.
- After at least one manual run and one scheduled hourly run produce expected evidence, the Codex app fallback can be paused or deleted.
Tasks
T01 - Confirm Runtime Substrate And Cutover Boundary
id: CUST-WP-0046-T01
status: done
priority: high
state_hub_task_id: "99e16a50-9775-4520-b343-f52fda0b67ec"
Confirm the activity-core production deployment on Railiance01 is usable as the primary scheduler before adding more automation.
Checks:
- activity-core API health endpoint reachable from the operator path
- worker running and connected to Temporal
- Temporal UI or CLI can list schedules and recent workflows
- State Hub URL and credentials/environment are available to activity-core
- llm-connect is not required for this RecentlyOnScope path
- current Codex automation fallback status is recorded before any change
Done when the hourly RecentlyOnScope rollout has a verified activity-core host and a written cutover boundary: Codex remains fallback until T06, then is paused/deleted.
T02 - Add State Hub Active-Domain Batch Generation
id: CUST-WP-0046-T02
status: done
priority: high
depends_on: [CUST-WP-0046-T01]
state_hub_task_id: "c5004c0b-a261-407c-9376-a33883e054bf"
Extend State Hub's existing RecentlyOnScope functionality instead of creating a parallel generator in activity-core.
Required behavior:
- accept a range such as
1h, defaulting to1h - compute the window in UTC
- select active domains with at least one qualifying source in the window: progress events, decisions, updated workstreams, or updated tasks
- optionally include domains with open human-intervention items when configured
- generate one RecentlyOnScope report per active domain using the existing
generate_report()service - return metadata for generated/skipped/failed domains
- log one State Hub progress event with event_type
recently_on_scope_hourly
Proposed endpoint:
POST /recently-on-scope/hourly
Done when State Hub tests cover active-domain detection, no-active-domain behavior, report idempotency, partial failures, and report metadata response.
T03 - Add activity-core State Hub Batch Invocation Capability
id: CUST-WP-0046-T03
status: done
priority: high
depends_on: [CUST-WP-0046-T02]
state_hub_task_id: "a33b379f-56ba-47d0-b73c-3e724b1c5d45"
Add the smallest reusable activity-core capability needed to invoke the State Hub batch endpoint from an ActivityDefinition.
Preferred implementation:
- a deterministic State Hub action activity or sink, not an LLM instruction
- bounded timeout and retry policy
- response metadata persisted into the ActivityRun audit trail
- clear failure behavior: failed batch run marks the Temporal workflow failed or emits a visible blocked/error progress event
- no domain-selection logic inside activity-core
Done when a synthetic ActivityDefinition can call a mocked State Hub batch endpoint and record the result under test.
T04 - Create Hourly RecentlyOnScope ActivityDefinition
id: CUST-WP-0046-T04
status: done
priority: high
depends_on: [CUST-WP-0046-T03]
state_hub_task_id: "dcb20f5a-c446-48d6-b810-84de365c22fd"
Create a domain-owned ActivityDefinition in
activity-definitions/hourly-recently-on-scope.md.
Expected definition:
- trigger: cron expression for hourly execution, timezone
Europe/Berlin - misfire policy:
skip - enabled:
falseuntil manual canary passes - action: call State Hub RecentlyOnScope hourly batch endpoint with range
1h - report/audit sink: State Hub progress event and activity-core run metadata
- no LLM model configuration
- clear note that this is the first owned routine replacing Codex fallback habits
Done when the definition parses, syncs into activity-core, and appears as a paused Temporal schedule while disabled.
T05 - Manual Canary And Scheduled Canary
id: CUST-WP-0046-T05
status: done
priority: high
depends_on: [CUST-WP-0046-T04]
state_hub_task_id: "f5c0cf64-a8e9-4d8c-bd86-ca58cbf132c2"
Validate the hourly routine before enabling it as the standing habit.
Manual canary evidence:
- manual trigger workflow id
- ActivityRun row
- State Hub progress event with event_type
recently_on_scope_hourly - generated report paths for every active domain
- no generated reports for inactive domains unless configured
- no duplicate report when rerun for the same window
Scheduled canary evidence:
- hourly schedule unpaused only after manual canary passes
- next hourly run completes from activity-core
- generated report metadata matches the active-domain window
- missed-run behavior is confirmed as
skip
Done when both manual and scheduled canaries leave complete evidence.
T06 - Retire Codex Automation Fallback
id: CUST-WP-0046-T06
status: done
priority: high
depends_on: [CUST-WP-0046-T05]
state_hub_task_id: "2a46a6c8-4d3e-4064-a935-c90ca0c76a6d"
Remove the Codex app automation fallback from the operating path.
Steps:
- decide whether to pause or delete
daily-state-hub-wsjf-triage - record the decision in State Hub
- apply the Codex automation change
- update CUST-WP-0045 notes so the daily triage handoff no longer points at an active Codex fallback
- ensure there is exactly one active scheduled substrate for routine reports: activity-core
Done when Codex app automation no longer runs routine Custodian reporting and the owned activity-core schedule has verified evidence.
T07 - Observability, Runbook, And Retention
id: CUST-WP-0046-T07
status: done
priority: medium
depends_on: [CUST-WP-0046-T05]
state_hub_task_id: "01066ff8-f591-43f6-99e6-d2f61654e590"
Document how operators answer "did the hourly report run?" without opening Codex Desktop.
Runbook should include:
- activity-core schedule check
- Temporal workflow query
- ActivityRun query
- State Hub progress query for
recently_on_scope_hourly - RecentlyOnScope report list endpoint
- report directory and retention expectations
- behavior when activity-core host is offline at the top of the hour
Done when the operator can verify hourly report health from State Hub and activity-core telemetry alone.
Implementation Evidence - 2026-05-22
Implemented pieces:
- State Hub now exposes
POST /recently-on-scope/hourly. - State Hub batch generation reuses the existing RecentlyOnScope collector, renderer, report id, and report directory.
- The batch endpoint selects domains by qualifying activity in the requested window: progress events, decisions, updated workstreams, or updated tasks.
- Domains with only registered repositories are skipped; domains with open
human-intervention items can be included by setting
include_attention: true. - The batch endpoint records one
recently_on_scope_hourlyprogress event with generated, skipped, and failed domain metadata. - activity-core has a reusable State Hub resolver query
recently_on_scope_hourlythat POSTs to the batch endpoint. - activity-core context sources can now be marked
required: true; required resolver failures fail the workflow instead of silently binding{}. - Custodian now owns
activity-definitions/hourly-recently-on-scope.md, disabled until canary. - Operator verification notes live in
docs/hourly-recently-on-scope-runbook.md.
Verification:
- State Hub:
/home/worsch/.local/bin/uv run pytest tests/test_recently_on_scope.pypassed with 11 tests. - State Hub:
/home/worsch/.local/bin/uv run pytest tests/test_recently_on_scope.py tests/test_mcp_smoke.py::TestAddProgressEventpassed with 14 tests. - activity-core:
/home/worsch/.local/bin/uv run pytest tests/test_state_hub_context_resolver.py tests/test_sync_activity_definitions.py tests/test_schedule_lifecycle.pypassed with 19 tests. - activity-core parser scan with
ACTIVITY_DEFINITION_DIRS=/home/worsch/the-custodianfoundHourly RecentlyOnScope Reports.
Local canary evidence:
- Direct State Hub batch canary generated reports for
custodianandmarkitect, skipped 9 quiet active domains, and had no failed domains. Progress event:ff1e845f-df02-476e-a3f3-fb912e4f32a8. - Direct activity-core resolver canary invoked the same State Hub batch endpoint
and generated reports for
custodianandmarkitect, skipped 9 domains, and had no failed domains. Progress event:1ff091c3-227c-4a85-b5e2-172ba45a2676.
Railiance01 cutover evidence - 2026-05-23:
- Runtime substrate verified: activity-core API health returned healthy DB and Temporal status, worker and API deployments rolled out, and Temporal CLI can list schedules/workflows from the Railiance01 cluster.
- Live State Hub access now uses an in-cluster bridge service
actcore-state-hub-bridgebecause the State Hub tunnel on Railiance01 is bound to node-local127.0.0.1:18000. The activity-core runtime config now pointsSTATE_HUB_URLathttp://actcore-state-hub-bridge:8000. - The Railiance01 manifest mounts
actcore-external-activity-definitions/hourly-recently-on-scope.mdinto the sync, API, and worker pods. Sync logs showHourly RecentlyOnScope Reportsupserted into the live activity-core DB. - The first live manual attempts recorded
{}because the worker image still had the old resolver behavior and the runtime URL was pointed at the wrong service. activity-core now validates that the hourly State Hub response containsgenerated,skipped, andfailedbefore accepting the run. - Manual canary after image rollout:
workflow
activity-d104348c-d792-4377-943c-70a31e81a9bc:manual-0be193ee-341d-441b-9bc7-e1696b96455b, ActivityRun43070cac-2669-5e53-bb9e-88dc1000f3a1, progress event96bf2f6b-38e3-4143-a49b-1b1275a9713a, one generatedcustodianreport, ten skipped quiet domains, and no failed domains. - Scheduled canary via Temporal
schedule trigger: workflowactivity-d104348c-d792-4377-943c-70a31e81a9bc:${firstScheduledTime}-2026-05-23T00:20:30Z, ActivityRunef741a08-3e21-5945-ae21-fdf9b1968438, progress event8d284579-5652-4c6e-8783-36fde75f8ed4, no generated reports for the quiet window, eleven skipped quiet domains, and no failed domains. - The hourly Temporal schedule is unpaused with
misfire_policy: skip; the API reports the ActivityDefinitionenabled: true.
Remaining gate:
- T06 remains blocked because
daily-state-hub-wsjf-triageis a daily WSJF triage fallback, not the hourly RecentlyOnScope runner. The daily activity-core definition is stillenabled: false, so pausing or deleting the Codex app automation here would remove daily WSJF coverage beforeCUST-WP-0045has its own canary.
This gate cleared on 2026-06-04 after CUST-WP-0045 finished its activity-core
daily WSJF cutover and CUST-WP-0044 finished the three-run calibration.
Implementation Evidence - 2026-06-04
T06 is done.
- Decision: keep the old Codex Desktop automation
daily-state-hub-wsjf-triagepaused rather than deleting it immediately. This preserves a recoverable fallback without allowing it to run as a second routine scheduler. - Evidence: local Codex automation metadata shows
daily-state-hub-wsjf-triagewithstatus = "PAUSED". activity-definitions/daily-statehub-wsjf-triage.mdisenabled: trueand now names activity-core as the active owned runner.CUST-WP-0045is finished; its T08 calibration used the June 2-4 activity-core daily triage notes.- The 2026-06-04 calibration also updated the daily triage prompt/schema so future reports include explicit WSJF ranks and component scores.
- Result: routine Custodian reporting no longer depends on an active Codex app
automation fallback. Daily WSJF and hourly RecentlyOnScope now both have
owned activity-core paths, so
CUST-WP-0046is finished.
Acceptance Criteria
- Hourly RecentlyOnScope reports are generated by activity-core, not Codex app automation.
- State Hub owns active-domain selection and report generation.
- Activity-core owns scheduling, run history, and failure visibility.
- The run uses existing RecentlyOnScope report storage and API surfaces.
- A manual canary and a scheduled canary both pass.
- Codex app automation fallback is paused or deleted after activity-core is verified.
- No LLM provider is required for the hourly RecentlyOnScope routine.
- Missed runs are skipped rather than replayed in a burst.
Notes
This work complements CUST-WP-0045. Daily WSJF triage still needs its own
LLM-backed canary before that specific report can be enabled in activity-core,
but RecentlyOnScope is deterministic and can become the first always-on
activity-core reporting habit sooner.