# Current Account Dataset Review

> Review and repair current-account TianGong dataset snapshots across flow, process, and lifecyclemodel rows through public CLI commands only. Use when a task needs local schema validation, deterministic reference rewrites, lifecyclemodel save-draft handoff, lifecyclemodel graph audit, and artifact-first stop rules for account-owned dataset governance.

- Skill: `tiangong-lca/current-account-dataset-review` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add tiangong-lca/current-account-dataset-review`
- Raw SKILL.md: https://api.skillmd.com/api/skills/tiangong-lca/current-account-dataset-review/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: tiangong-lca (https://skillmd.com/u/tiangong-lca)
- Updated: 2026-09-21
- Page: https://skillmd.com/skills/tiangong-lca/current-account-dataset-review

---


# Current Account Dataset Review

## Scope

- This skill is a thin orchestration layer over public `tiangong` CLI commands.
- Use it for current-account multi-type dataset governance when the input rows or manifests are already frozen locally.
- Do not import `tiangong-cli/dist/src/lib/**`, write direct Supabase REST glue, or parse raw credentials in this skill.
- If a capability is missing from the public CLI, record it as a CLI gap instead of adding private runtime logic here.

## Canonical Flow

1. Before remote account work, run `tiangong-lca auth status --json`. If it returns `login-required`, stop and ask the human to run `tiangong-lca auth login` in a trusted terminal. Run `tiangong-lca auth doctor-auth --json` before a commit or account-sensitive readback. Never ask for a username, password, authorization code, token, or legacy API key.
2. Freeze or receive the account-scoped input rows.
3. Validate all local rows:

```bash
node current-account-dataset-review/scripts/run-current-account-dataset-review.mjs validate \
  --input /abs/path/rows.jsonl \
  --type auto \
  --out-dir /abs/path/dataset-validate
```

4. If an upstream flow version changes, rewrite downstream references locally first:

```bash
node current-account-dataset-review/scripts/run-current-account-dataset-review.mjs rewrite-references \
  --input /abs/path/rows.jsonl \
  --from flow:<old-id>@<old-version> \
  --to flow:<new-id>@<new-version> \
  --type process \
  --type lifecyclemodel \
  --out-dir /abs/path/reference-rewrite \
  --dry-run
```

5. Refresh references against reachable remote rows when the task requires current published versions:

```bash
node current-account-dataset-review/scripts/run-current-account-dataset-review.mjs refresh-remote-references \
  --input /abs/path/rows.jsonl \
  --out /abs/path/rows.refreshed.jsonl \
  --out-dir /abs/path/reference-refresh \
  --dry-run
```

Input artifact: frozen flow, process, or lifecyclemodel rows.
Output artifacts: refreshed rows, remote lookup report, and blockers emitted by `tiangong-lca dataset references refresh-remote`.

6. Save lifecyclemodel drafts only after local validation passes:

```bash
node current-account-dataset-review/scripts/run-current-account-dataset-review.mjs save-lifecyclemodels \
  --input /abs/path/lifecyclemodels.jsonl \
  --out-dir /abs/path/lifecyclemodel-save-draft \
  --dry-run
```

7. Generate graph artifacts and connection findings for lifecyclemodels:

```bash
node current-account-dataset-review/scripts/run-current-account-dataset-review.mjs graph-lifecyclemodels \
  --input /abs/path/lifecyclemodels.jsonl \
  --out-dir /abs/path/lifecyclemodel-graph \
  --format all \
  --check-connections
```

8. After any remote write, re-fetch or freeze the persisted rows and run remote/reference verification:

```bash
node current-account-dataset-review/scripts/run-current-account-dataset-review.mjs verify-remote \
  --input /abs/path/rows.jsonl \
  --out-dir /abs/path/dataset-verify-remote
```

Input artifact: frozen persisted rows or row snapshots.
Output artifact: `dataset-remote-verify` report from `tiangong-lca dataset verify-remote`.

## Runtime Contract

- Wrapper-local `--cli-dir` is the only supported override for choosing a local CLI checkout.
- Local dry-runs do not require remote credentials.
- Commit paths use the canonical CLI env only:
  - `TIANGONG_LCA_API_BASE_URL`
  - `TIANGONG_LCA_SUPABASE_PUBLISHABLE_KEY`
  - `TIANGONG_LCA_OAUTH_CLIENT_ID`
- Use a separate private `TIANGONG_LCA_SESSION_FILE` for each account/project/client. Headless runs may use only an orchestrator-injected short-lived `TIANGONG_LCA_ACCESS_TOKEN` that never enters argv, prompts, logs, or artifacts.

## Stop Rules

- Any commit must be preceded by a dry-run artifact set.
- If a publish or save-draft step partially fails, do not rerun the full scope blindly; use the generated failure artifacts to plan a targeted retry.
- After changing flow or process versions, re-validate downstream process and lifecyclemodel references.
- Lifecyclemodel review must include schema validation and graph connection checks.
- Final completion for remote write tasks requires a fresh `dataset verify-remote` pass.
- If any upstream identity gate reports `manual_review` or `block_duplicate`, stop account-level mutation and preserve the gate artifact path in the handoff.
- Treat schema, reference-refresh, remote-verification, graph, and post-write verification blockers as hard stops.

## Known CLI Gaps

- `dataset inventory --current-user`
- `flow publish-version` server-side next available version / conflict retry
- `dataset scope expand`