repo-manager/workplans/RMGR-WP-0008-work-record-and-register-receiving-surface.md
tegwick d103955217 feat: finish register receiving and authority routing
Assistant: codex
Assistant-Model: gpt-5.6-sol
Assistant-Session: 01a023c0-a0a3-7c03-b395-5a0d2757214d
2026-08-21 23:15:48 +02:00

10 KiB

id type title domain repo status owner topic_slug created updated parent_project parent_workplan related state_hub_workstream_id
RMGR-WP-0008 workplan Receiving surface for work records and repository registers infotech repo-manager finished codex infotech 2026-08-20 2026-08-21 prj-state-hub-retirement SHR-WP-0001
STATE-WP-0079
RMGR-WP-0002
RMGR-WP-0003
RMGR-WP-0004
HUB-WP-0004
f45cebf3-7d49-471d-a903-d711e19f7653

Receiving surface for work records and repository registers

Goal

Build the repo-manager surfaces that STATE-WP-0079's cutover slices need in order to move. Today most of those slices have nowhere to land, and that — not anything inside State Hub — is what blocks the retirement.

Why this exists

STATE-WP-0079-T02 produced a cutover slice plan (state-hub/docs/retirement-cutover-slice-plan.md) assigning all 425 SHR-INV-0001 items to slices. Its first owner-readiness pass read readiness off workplan status: RMGR-WP-0001/0002/0003 are finished, so all 167 repo-manager-owned items were marked receivable.

That was wrong, and the correction (2026-08-20) is the reason for this workplan. Verified against src/repo_manager/:

  • parse/ contains only workplan.py
  • rmgr subcommands are observe, reconcile, task-status, conform, prefix, scaffold, dual-run-status
  • the source tree has no reference to sbom, repo_goal, contribution, technical_debt, or extension_point
  • the State Hub adapter calls exactly three operations — rm_update_task_status, rm_reconcile, rm_scaffold

What RMGR-WP-0001/0002/0003 delivered is work-record dual-run for task status, repository observation, and scaffolding. Not a general work-record write path, and not a register surface at all.

Net effect: of 425 inventory items, roughly 6 can move today (slice A2a, task status + reconcile). The rest wait on surfaces built here or in hub-core.

Scope boundary

This workplan builds receiving surfaces in repo-manager. It does not change State Hub beyond the adapter calls that route to these surfaces.

Building any of this in State Hub instead is inadmissible under state-hub/policies/retirement-freeze.md: it would establish new permanent ownership in the component being retired, and would build the capability twice.

Per SCOPE.md, file-backed records in domain repos stay authoritative; these surfaces index and govern them, they do not become a second source of truth.

Workplan write path

id: RMGR-WP-0008-T01
status: done
priority: high
state_hub_task_id: "5ef517b4-94a6-48d4-b200-827460f79e92"

Extend the work-record write path beyond task status to workplan create, update, status change, and delete. parse/workplan.py already reads workplans; this adds the governed mutation side and an rm_update_workplan counterpart to rm_update_task_status.

Unblocks slice A2b (~9 routes plus table:workplans).

Follow the RMGR-WP-0002 dual-run shape — writeback, reconcile, mutation meter — rather than inventing a second mechanism.

Result (2026-08-21): rmgr workplan create|update|delete now owns the file-backed mutation path. Update accepts canonical ids or State Hub UUIDs; delete is a confirmed, recoverable archive operation. All three operations use optimistic HEAD checks, idempotency, Git evidence, reindexing, normalized events, mutation metering, and optional push-seal. State Hub has the matching rm_update_workplan compatibility adapter. Repo Manager's full suite and the adapter's focused compatibility tests pass.

Intake records

id: RMGR-WP-0008-T02
status: done
priority: medium
state_hub_task_id: "017f3b29-7100-4771-98c0-59fe91895a5f"

Add an intake parser and index entry kind, plus the write path for create, route, note, and close. Intake is the successor to the retired suggestion backlog, so this is the discovery surface's permanent home.

Unblocks slice A2c (9 items: 7 routes, table:intakes, table:intake_notes).

Result (2026-08-21): Repo Manager indexes typed intake YAML fences from repository record locations and provides governed intake create|route|note|close commands. Mutations are file-backed, committed with correlation evidence, idempotent when keyed, reindexed, and metered.

Decision records

id: RMGR-WP-0008-T03
status: done
priority: medium
state_hub_task_id: "392ac64d-15b6-4f4e-9e18-96dd7d299208"

Add a decision parser, index entry kind, and write path including resolve.

Note the coupling: resolve_decision is one of only two write operations sanctioned against the hub in the fleet agent instructions. Whatever replaces it must keep that guarantee — agents rely on it being available and safe.

Unblocks slice A2d (8 items, including caller:agentic-resources).

Result (2026-08-21): typed decision records share the Markdown/YAML receiving spine and expose governed decision create|update|resolve. Resolve refuses incomplete decisions unless both rationale and deciding identity are present, preserving the safety guarantee of the former hub operation.

Dependencies and human-flag records

id: RMGR-WP-0008-T04
status: done
priority: low
state_hub_task_id: "216a03e6-a0ee-4e30-ae96-fa11e77a22ac"

Workplan dependency edges and human-intervention flags. Smallest of the work-record slices and the least coupled; a reasonable first cut if T01 proves too large to land in one piece.

Unblocks slice A2e (~6 items) and the dependency routes inside A2b.

Result (2026-08-21): workplan and task depends_on values, plus task needs_human, intervention_note, and blocking_reason, survive parsing into the replaceable index. Their authority remains the workplan file.

Repository registers

id: RMGR-WP-0008-T05
status: done
priority: high
state_hub_task_id: "902c71da-1ba5-4e76-bf20-0d9f697edf7a"

The largest gap. Six registers with no counterpart in repo-manager at all:

Register Items Notes
sbom-inventory 10 lockfile ingest, licence report, copyleft detection
repo-goals 9
upstream-contributions 9
register-technical-debt 9 also backs dashboard UI feedback — check before moving
register-extension-points 6
register-entries 6 MCP tools only

Unblocks slice A3 (49 items).

Treat as one register spine with per-kind schemas rather than six unrelated features — they share repo scoping, append-mostly write patterns, and the same file-backed-authority rule.

Result (2026-08-21): the shared spine is live for all six kinds as versioned registers/<kind>.yaml files, with parser/index support and governed put/defer/note commands. Kind-specific required fields cover the existing State Hub create contracts, and the generic rm_update_register_entry adapter can route compatibility writes. rmgr sbom scan|licence-report derives versioned, provenance-bearing snapshots from current lockfiles (plus a reviewed tool manifest), with direct-production copyleft findings. The State Hub adapter and route-level fixture cover the read/derive call.

The 22 historical State Hub snapshots were reconciled rather than copied: the newest projection for each of 18 repositories was compared with a fresh scan, all repositories parsed, and 9 counts matched exactly. Current lockfiles yield 1,565 entries versus 2,659 in months-old projections. Copying stale cache rows would invert repository authority, so cutover rebuilds from source and retains old snapshot identifiers only as retirement provenance. Evidence: docs/evidence/RMGR-WP-0008-sbom-cutover-2026-08-21.md.

Topic and classification contract

id: RMGR-WP-0008-T06
status: done
priority: medium
state_hub_task_id: "ef7934de-a1e2-4541-8068-e75876446b60"

observe/load_classification read .repo-classification.yaml today. Slice A5 (topic-spine, 8 items) is dispositioned replace, so it needs a published contract rather than a lift-and-shift of the hub's tables.

Coordinate with HUB-WP-0004: the topic spine sits on the boundary between repository classification (here) and cross-domain projection (hub-core).

Result (2026-08-21): published docs/repository-classification-contract_v1.0.{md,yaml} and made observation reject invalid controlled values, repeated primary domains, malformed lists, and non-kebab capability tags. Finished HUB-WP-0004 explicitly accepts that repository classification remains authoritative here while hub-core owns cross-domain projections. That is the required boundary agreement; consumer route switching remains in the existing STATE-WP-0079 retirement workplan.

Owner-side UI

id: RMGR-WP-0008-T07
status: done
priority: low
state_hub_task_id: "24ee329d-85dd-4e11-9c87-2f20b8f86e81"

Slice A4 (22 items, replace) assumes an owner-side dashboard. rmgr is a CLI; there is none, and it is not obvious there should be one — the work-coordination and repo views may belong to hub-core's projection layer instead.

Resolved in docs/adr-003-projection-ui-ownership.md: Repo Manager owns its governed CLI/API contracts, not a dashboard. hub-core owns cross-repository projection/query and surviving hub UI capabilities; ops/domain UIs are clients of those contracts. A4 is therefore a hub-core/projection-client replacement, not a UI migration into Repo Manager.

Acceptance

  • Workplan write path live, A2b routable through the adapter
  • Intake and decision parsers + write paths live; resolve_decision guarantee preserved
  • Register spine live, A3's six registers routable
  • Topic/classification contract published and accepted at the hub-core boundary
  • A4 ownership decided and recorded
  • Compatibility tests per slice, matching the RMGR-WP-0002 evidence pattern

Completion and handoff

Repo Manager's receiving surfaces are complete. The remaining State Hub route flags, redirects, projection rebuilds, retention, and destructive retirement gates already remain live in STATE-WP-0079; they are consumers of this work, not residual Repo Manager implementation. No actionable Repo Manager residual is being parked in this finished workplan.