rein-aharness/workplans/REIN-A-0002-ops-run-claim-loop.md

183 lines
5.4 KiB
Markdown

---
id: REIN-A-0002
type: workplan
title: "Ops run claim loop and approach selection"
domain: agents
repo: rein-aharness
status: finished
owner: grok
topic_slug: rein-aharness
priority: high
created: "2026-08-03"
updated: "2026-08-08"
depends_on:
- ACTIVITY-WP-0026
related:
- ACT-ADR-005
- HARNESS-WP-0001
state_hub_workstream_id: "ed679a3e-b4ea-4e69-bba6-5694d79ae7c8"
---
# REIN-A-0002 — Ops run claim loop and approach selection
## Origin
ACT-ADR-005. Closes the dual-clock residual: host systemd timers currently
execute Binky/FI while activity-core schedules separately. rein-aharness already
polls **issue-core** (`docs/task-intake.md`); that path is wrong for internal
ops under WP-0022 and issue-core INTENT.
## Goal
1. Claim **`ops_run`** from activity-core (not issue-core / not Forgejo).
2. Select an **efficient approach** (adapter table) and execute.
3. Complete/fail the ops_run + domain completion events.
4. Cut over FI and Binky off dual wall-clock timers once proven.
## Tasks
### T01 — activity-core ops_run client
```task
id: REIN-A-0002-T01
status: done
priority: high
state_hub_task_id: "a5e32652-ba3f-4cc5-8d41-0d9f3022c50e"
```
Client module (env: `ACTIVITY_CORE_URL`, worker credential):
- `claim(worker_id, labels=…)`
- `heartbeat` / `complete` / `fail`
Map ops_run → existing `TaskSpec` (or thin OpsRunSpec). Tests with httpx mock.
**Blocked on:** ACTIVITY-WP-0026-T01 API shape (can stub against openapi draft).
**Done when:** unit tests green; CLI `poll --source=ops-run` prototype optional.
---
### T02 — Approach registry (v1 static)
```task
id: REIN-A-0002-T02
status: done
priority: high
state_hub_task_id: "c307b119-1bff-4d5c-a911-5b5259de5c86"
```
Central table/config (code registry, not per-repo cron):
| Match (labels / definition id contains) | Approach command |
| --------------------------------------- | ---------------- |
| `research-brief` or `fi-daily` | `fi-research-brief` |
| `rhythm` + binky / `binky-daily` | `brief-daily` |
| `mail-intake` | `mail-scan` then optional `mail-triage` |
| `agent-session` | `run` with persona/profile |
| unmatched | fail ops_run with clear reason |
Document how to add rows. Prefer **cheapest correct** path (deterministic before
agent session; structured llm-connect before open-ended coding agent).
**Done when:** pure function tests for match → approach.
---
### T03 — Continuous claim worker
```task
id: REIN-A-0002-T03
status: done
priority: high
state_hub_task_id: "31fb7df3-83a3-4f18-a7b9-9be49e226837"
```
- `rein-aharness claim-loop` (or systemd/service unit): poll interval, lease
heartbeat, concurrency 1 default.
- On success: domain completion already done by approach; close ops_run.
- On failure: fail ops_run; do not silent drop.
- Metrics/log for SLA.
**Done when:** can run against railiance actcore-api with a test ops_run.
---
### T04 — Map FI + Binky definitions
```task
id: REIN-A-0002-T04
status: done
priority: high
state_hub_task_id: "93190150-4df7-47f4-b52e-29b997f9aaf0"
```
- FI labels already include `research-brief` + `automated`.
- `AGENT_HARNESS_REPO_MAP` documented in claim-loop.env.example + ops-run-claim-loop.md.
- Install: `deploy/scripts/install-claim-loop-user.sh`.
**Prod e2e 2026-08-03:** FI trigger → ops_run claimed by
`rein-aharness@railiance01` → approach `fi-research-brief`**succeeded**
(skipped existing brief). User unit `rein-aharness-claim-loop.service` active.
Proof window for timer cutover (T05) starts here.
---
### T05 — Cutover: timers become break-glass
```task
id: REIN-A-0002-T05
status: done
priority: medium
state_hub_task_id: "a04a98e3-ca6a-4a18-b6fd-47a77047ca90"
```
**Procedure landed:** `deploy/docs/timer-cutover.md` + FI `docs/recurrence-ops.md`
updated for claim-loop primary / timer break-glass.
Completed on railiance (verified 2026-08-08):
1. Five-cycle proof table filled from production claim-loop journal (eight clean
FI/Binky cycles observed in total).
2. FI host timer is disabled/inactive; Binky host timer is not installed and
inactive.
3. Install scripts and manual service starts remain documented for break-glass.
**Done when:** docs + railiance state match; dual-clock residual closed for FI/Binky.
---
### T06 — Deprecate issue-core as ops intake for automation
```task
id: REIN-A-0002-T06
status: done
priority: medium
state_hub_task_id: "cca10835-d6fc-4ce1-a361-d2d1f2693872"
```
- `docs/task-intake.md`: issue-core poll = **legacy / external ticket path only**.
- ops_run = **primary** for activity-core automation labels.
- No Gitea references; Forgejo only if mentioning self-hosted forge projection.
**Done when:** docs and CLI help text consistent with ACT-ADR-005.
## Acceptance
- [x] Claim loop does not require Forgejo or issue-core for FI/Binky ops
- [x] Approach table selects fi-research-brief / brief-daily without host “schedule”
- [x] Timers demoted after proof window (**T05**)
- [x] ACT-ADR-005 / ACTIVITY-WP-0026 contracts implemented
**Shipped:** `ops_run_client`, `approaches`, `claim_loop`, CLI `claim-loop` /
`run --from-ops-run` / `poll --source=ops-run`, unit tests, install + cutover docs.
**Residual handoff:** `REIN-A-0003` captures the separately observed unmatched
Binky weekly-review ops run; it does not reopen the completed FI/Binky daily
cutover scope.
## Out of scope
- Implementing ops_run schema (ACTIVITY-WP-0026)
- issue-core Forgejo backend fixes
- Full glas-harness multi-rein routing