external-dataset-curated-import

name	external-dataset-curated-import
description	Orchestrate structured external LCA packages such as BAFU or USLCI into TIDAS candidate rows through CLI conversion, queue state, child skills, deterministic gates, and readback-safe publish handoff.

External Dataset Curated Import

Use this top-level skill when the input is a packaged or semi-structured LCA dataset that should enter TianGong through a repeatable import profile.

Boundaries

• This skill is a workflow coordinator only.
• Do not parse source package formats in this skill. Use tiangong-lca dataset import-lca convert or tidas-tools.
• Do not implement queue state, schema validation, QA, remote writes, or database verification in this skill. Use the CLI.
• Do not make BAFU, USLCI, or any account a code path. They are profile inputs.
• Do not hand-edit row JSONL as the primary repair path. Use child skills to write structured decisions or patches, then apply them through deterministic CLI/Foundry commands.

Required Inputs

• Foundry task directory under .foundry/workspaces/<task-id>/.
• Frozen source package manifest and checksum.
• Import profile id, such as generic, bafu, or uslci.
• Account/write guard when the task may eventually request remote write.
• Target entity scope: support, flow, process, lifecyclemodel, or a bounded subset.

Workflow

1. Fetch SDK-backed contract context for every target TIDAS type:

tiangong-lca dataset context-pack \
  --type <process|flow|source|contact|unitgroup|flowproperty|lifecyclemodel> \
  --profile ai-import \
  --out-dir .foundry/workspaces/<task-id>/context/<type> \
  --json

2. Convert the source package:

tiangong-lca dataset import-lca convert \
  --input /abs/path/source-package \
  --output-dir .foundry/workspaces/<task-id>/conversion \
  --from-format auto \
  --target tidas \
  --json

3. Normalize converted rows into explicit support, flow, and process JSONL files.

4. Run schema validation and deterministic QA through the CLI.

5. Build the entity queue:

tiangong-lca dataset curation-queue build \
  --processes .foundry/workspaces/<task-id>/rows/processes.jsonl \
  --flows .foundry/workspaces/<task-id>/rows/flows.jsonl \
  --support .foundry/workspaces/<task-id>/rows/sources.jsonl \
  --out-dir .foundry/workspaces/<task-id>/curation-queue \
  --json

6. Drive the queue with CLI state, not ad hoc batches:

tiangong-lca dataset curation-queue next \
  --queue-dir .foundry/workspaces/<task-id>/curation-queue \
  --json

Execute only the returned task. Support tasks complete before flow tasks; flow tasks complete before dependent process tasks.

7. For semantic blockers, route to existing child skills:

• Support/source/contact repair: use Foundry authoring task packages or dedicated support child skills when they exist.
• Flow repair: use flow-governance-review or a future flow-authoring alias.
• Process repair: use process-automated-builder or a future process-authoring alias.
• Publish handoff: use lca-publish-executor only after deterministic prewrite gates are ready.

8. After each task finishes, write or update that task's checkpoint.json, then call curation-queue next again.

9. Before remote write, verify queue completion for the intended scope:

tiangong-lca dataset curation-queue verify \
  --queue-dir .foundry/workspaces/<task-id>/curation-queue \
  --type process \
  --json

The verify report must be passed for the committed scope. If verify is blocked while next still returns ready, continue the queue.

10. Before the first remote account read or write, run tiangong-lca auth status --json; before commit, require tiangong-lca auth doctor-auth --json. If login is required, stop and ask the human to run auth login in a trusted terminal. Never request a username, password, authorization code, token, or legacy API key. Use a separate private session per account/project/client; headless tokens may come only from an approved orchestrator and never enter argv or artifacts.

Stop Rules

Stop and return blocker evidence when:

• source manifest or profile lock is missing;
• conversion reports blockers or unsupported source format;
• schema, QA, curation, or queue verify blocks with no runnable next task;
• a child skill would need to invent source evidence;
• remote write would require direct database access or bypass a CLI dry-run/verification gate.

Done Criteria

• Source package, profile lock, contract context, converted rows, schema/QA reports, queue artifacts, task checkpoints, mapping/provenance, dry-run, commit handoff, and readback verification are all present for the final scope.
• tiangong-lca dataset curation-queue verify is passed for the committed scope.
• Foundry task completion report aggregates the exact closeout artifacts.