reuse-surface/workplans/archived/260617-REUSE-WP-0013-registry-establish-and-llm-assist.md

id, type, title, domain, repo, status, owner, topic_slug, created, updated, state_hub_workstream_id

id	type	title	domain	repo	status	owner	topic_slug	created	updated	state_hub_workstream_id
REUSE-WP-0013	workplan	Registry establish, update, and stats with optional llm-connect assist	helix_forge	reuse-surface	finished	codex	helix-forge	2026-06-16	2026-06-17	239a0077-8593-4dc7-918d-4c23895275f6

Registry establish, update, and stats with optional llm-connect assist

Follow-up to operator feedback and REUSE-WP-0012-T01 blocks (history/2026-06-16-hub-registration-blocks.md). Sibling repos cannot federate until each publishes registry/indexes/capabilities.yaml; today that requires manual layout copy and authoring. This workplan adds deterministic bootstrap and analytics CLI plus optional llm-connect-backed discovery and refresh so domain repos can establish and maintain registries with low ceremony.

Baseline vector: D5 / A4 / C5 / R3
Target vector after completion: D5 / A4 / C5 / R3 (tooling depth; no product-vector promotion unless dogfood evidence warrants)

Problem statement

Pain	Today	Target
Bootstrap registry in sibling repo	Manual copy from reuse-surface	reuse-surface establish
Keep entries aligned with repo reality	Manual edits + validate	reuse-surface update
Portfolio / federation readiness view	report cohorts, manual curls	reuse-surface stats
Draft capability metadata from repo context	Agent improvisation	llm-connect structured extract

Design principles

1. Deterministic first — stats, establish --scaffold, and establish --publish-check work without llm-connect or API keys.
2. LLM optional — establish --discover and update --suggest call llm-connect HTTP (POST /execute) or library adapter when configured.
3. Dry-run default — LLM and write paths require explicit --apply.
4. Validate gate — every --apply path ends with reuse-surface validate (or fails with schema errors).
5. Boundary — reuse-surface does not embed provider keys; llm-connect owns routing and credentials.

Suggested execution order

T01 stats (deterministic, unblocks observability)
  → T02 establish --scaffold (unblocks sibling publish path)
  → T03 establish --publish-check (verifies raw URL / federation readiness)
  → T04 llm-connect bridge + draft JSON schema
  → T05 establish --discover
  → T06 update (deterministic signals + optional LLM suggest)
  → T07 docs, tests, gap-analysis note

Dependencies

Dependency	Owner	Notes
llm-connect server or package	llm-connect	LLM_CONNECT_URL or editable install
Capability JSON schema	reuse-surface	schemas/capability.schema.yaml unchanged
Sibling repo apply	Domain owners	establish run in target repo checkout
Hub token	Operator	hub register remains separate post-publish

Proposed CLI surface

# Deterministic
reuse-surface stats
reuse-surface stats --format json --federation-ready
reuse-surface establish --scaffold --domain helix_forge [--path .]
reuse-surface establish --publish-check --raw-url <gitea-raw-url>

# LLM-assisted (optional backend)
export LLM_CONNECT_URL=http://127.0.0.1:8088
reuse-surface establish --discover [--path .] --dry-run
reuse-surface establish --discover --apply
reuse-surface update --capability <id> --dry-run
reuse-surface update --all --suggest-maturity --dry-run
reuse-surface update --from-git-since HEAD~5 --apply

---

Add Registry Stats Command

id: REUSE-WP-0013-T01
status: done
priority: high
state_hub_task_id: "98e65330-bfc7-4282-b372-d35542b899ce"

Deliver reuse-surface stats with deterministic aggregates:

• Capability count; maturity histogram (D/A/C/R bands)
• Entries at R0–R2 vs R3+; consumption mode counts
• Index vs entry vector drift count
• Federation readiness: local sources.yaml / index presence; optional --federation-ready raw URL probe (HTTP status)
• Hub summary when REUSE_SURFACE_URL set (registration count)

Output: Markdown default, --format json. Pytest coverage. Document in tools/README.md.

Implement establish --scaffold

id: REUSE-WP-0013-T02
status: done
priority: high
state_hub_task_id: "b8fedd87-d0d3-41b4-9af8-e36d52bfe1c5"

Bootstrap registry/ in target repo (--path, default cwd):

• registry/README.md (pointer to reuse-surface schema and validate)
• registry/capabilities/.gitkeep
• registry/indexes/capabilities.yaml with version, domain, capabilities: []
• Refuse overwrite unless --force
• Print next steps: add entry, validate, merge to main, publish-check

No llm-connect dependency. Pytest with temp directory.

Implement establish --publish-check

id: REUSE-WP-0013-T03
status: done
priority: medium
state_hub_task_id: "2924d685-709f-4e28-886f-b363cd9c40b4"

Federation publish helper for sibling repo operators:

• curl-equivalent probe of --raw-url (status, content-type hint)
• Validate local index YAML if --path has registry/indexes/capabilities.yaml
• Report pass/fail with remediation tied to docs/RegistryFederation.md
• Exit non-zero on hard failures (non-200, invalid YAML)

Add llm-connect Bridge And Draft Schema

id: REUSE-WP-0013-T04
status: done
priority: high
state_hub_task_id: "650ebee5-b34b-4ed8-891d-d93aacebadd7"

Thin client boundary:

• reuse_surface/llm_bridge.py — POST {LLM_CONNECT_URL}/execute, parse JSON from response content
• schemas/registry-draft.schema.json — structured draft shape (capability list with id, name, summary, vector, tags, consumption_modes, discovery intent)
• Env: LLM_CONNECT_URL (default none); graceful error when missing on LLM paths
• Optional pyproject.toml extra: llm → llm-connect dependency
• Pytest with mocked HTTP (no live LLM in CI)

Implement establish --discover

id: REUSE-WP-0013-T05
status: done
priority: medium
state_hub_task_id: "b9154889-f538-4266-9918-b277f9a297be"

LLM-assisted bootstrap after --scaffold or on empty registry:

• Collect context files: INTENT.md, SCOPE.md, README*, pyproject.toml, AGENTS.md, top-level package dirs (configurable --context-max-files)
• Prompt template + schema-constrained JSON via llm-connect
• --dry-run: print proposed entries and index rows
• --apply: write registry/capabilities/*.md from template merge + index update; run validate before success exit
• Document prompt assumptions and review checklist in registry/README.md

Implement update Command

id: REUSE-WP-0013-T06
status: done
priority: medium
state_hub_task_id: "b79558da-54b2-4712-91d2-b298c7cf2c40"

Refresh existing entries from repo signals:

Deterministic (no LLM):

• Index/entry vector mismatch detection
• New paths under tests/, CLI modules → suggest evidence.tests / availability.current_artifacts additions
• --apply for safe deterministic patches only (explicit list in code)

Optional LLM (--suggest, --suggest-maturity):

• Git diff or file snapshot → proposed promotion_history, evidence notes
• Always --dry-run unless --apply; validate after apply

Targets: single --capability, --all, --from-git-since <ref>.

Documentation, Tests, And Gap Note

id: REUSE-WP-0013-T07
status: done
priority: low
state_hub_task_id: "a55a2f26-004e-4c20-90cb-49bed64a1291"

• tools/README.md — full command reference
• docs/RegistryFederation.md — link establish/publish-check to sibling onboarding
• docs/IntentScopeGapAnalysis.md — add priority 24 (registry bootstrap tooling); mark open
• SCOPE.md — "What Is Possible Now" when T01–T03 ship (incremental)
• CI: stats informational step; no llm-connect in CI
• Total pytest increase; reuse-surface validate unchanged

---

Acceptance

• reuse-surface stats reports maturity and federation-readiness aggregates
• establish --scaffold creates valid empty registry layout without overwrite accidents
• establish --publish-check detects 303 vs 200 raw URL outcomes
• llm-connect bridge works with mocked HTTP; fails clearly when URL unset
• establish --discover --dry-run produces schema-valid draft JSON from fixture context
• update --dry-run reports deterministic drift on sample repo
• All new commands documented; gap priority 24 recorded

Completion notes (2026-06-17)

• Modules: stats.py, establish.py, registry_update.py, llm_bridge.py
• Schema: schemas/registry-draft.schema.json
• validate --root for sibling repo validation after establish --apply
• 43 pytest tests; optional pip install -e ".[llm]" extra

Out of scope

• Auto hub register (still operator step with token)
• Auto-merge LLM output without human review path
• Embedding similarity or overlap ML (keep overlaps token heuristic)
• llm-connect hosting or provider configuration inside reuse-surface

Dogfood target

Run establish --scaffold and establish --publish-check against state-hub checkout when available; optional establish --discover to seed capability.statehub.workstream-coordinate from existing docs.