aidp-fusion-status

# aidp-fusion-status — is my pipeline healthy, and what's stale? The read-only observability member of the family. It reports per-node health by **reconciling two truth sources** and never changes anything. ## When to use - "Did my last seed/incremental succeed? what failed?" - "What's stale / hasn't refreshed / needs re-running?" - "Which gold marts are actually materialized, and with how many rows?" - As a pre-check before `/aidp-fusion-seed` or `/aidp-fusion-incremental`. ## When NOT to use - To *run* anything — seeding is `/aidp-fusion-seed`, deltas are `/aidp-fusion-incremental`. This skill is strictly read-only. ## Evidence discipline (load-bearing) **The live table is the evidence; the state row is the overlay.** `fusion_autopilot_state` is run *metadata* — a `success` row can sit next to a dropped/empty mart, and a table can exist with no state row. So health keys off the **live AIDP catalog** (what's materialized), cross-checked against the state row. Never report "healthy" from a state row alone. (Same rule as the seed guard / the advisor — [[feedback_live_catalog_is_evidence]].) ## Helper | File | Role | Invoked via | |---|---|---| | `status_report.py` | Classify per-node health from `{state rows} × {live tables}`. | `Bash`, JSON in/out | Health classes: `HEALTHY` · `EMPTY_OK` · `STALE` (success recorded, table absent) · `FAILED` · `SKIPPED` · `UNTRACKED` (table exists, no run record) · `NEVER_RUN` (declared node, nothing anywhere). ## Workflow 1. **Acquire both signals** (cluster-side — the laptop `aidp-fusion-autopilot status` is local-Spark-only and has no JSON): ```bash .venv/bin/python tests/live/aidp_state_probe_live.py \ --aidp-id <OCID> --workspace-key <key> --cluster-key <key> \ --cluster-name <name> --out status.json ``` It runs, on an ACTIVE cluster, `SHOW TABLES` + `COUNT(*)` over the gold (and optionally silver/bronze) schema → `live`, and the latest-per-dataset query over `<catalog>.<bronzeSchema>.fusion_autopilot_state` → `state`, into one JSON. *(When `aidp-fusion-autopilot status --json` ships, use it instead.)* 2. **Classify**: ```bash python3 skills/aidp-fusion-status/status_report.py --input status.json ``` Optionally add `known_nodes` (from `content-pack info <pack> --json` + bundle datasets) so declared-but-never-built nodes show as `NEVER_RUN`. 3. **Render** a per-node ledger + the `attention` list (STALE/FAILED/UNTRACKED/ NEVER_RUN), with `lastRun` / `lastStatus` / `stateRowCount` vs `liveRowCount`. 4. **Recommend, don't act** — e.g. "re-run `gl_balance` (STALE) via `/aidp-fusion-seed`"; "investigate `mystery_mart` (UNTRACKED — built outside the orchestrator)"; "`po_backlog` NEVER_RUN — seed it or drop it from the bundle." Never run anything yourself. ## Output (example ledger) ``` fusion_catalog.gold — 4 nodes ✓ HEALTHY ar_invoice_summary last=2026-06-15 success state=49 live=49 ✓ HEALTHY supplier_spend last=2026-06-14 success state=212 live=212 ⚠ STALE gl_balance last=2026-06-09 success live=absent ⚠ UNTRACKED _phase5_probe — no run record (excluded: probe table) attention: gl_balance (re-seed), ``` ## Skill family - **Consumed by `aidp-fusion-autopilot`** as a state detector ("where am I / what's stale?"). - Pairs with `/aidp-fusion-seed` (first build) and the planned `/aidp-fusion-incremental` (deltas) — status tells you *what* to (re)run; they run it. ## Safety invariants - **Read-only** — never seeds, runs, or mutates anything. - **Live table = evidence**; never report HEALTHY off a state row alone. - Exclude audit/probe tables (`fusion_autopilot_state`, `_*`) from UNTRACKED noise.