rye-import-inspector

name	rye-import-inspector
description	Validate Rye-tracked import and update runs before or after target writes. Use when Codex needs deterministic checks for source_row, mapped_record, change-plan, metadata, Rye audit, old-value traceability, table declarations, approvals, touched IDs, or post-write verification evidence.

Rye Import Inspector

Use this skill as the generic validation gate for imports that rely on Rye for traceability. It checks process evidence; domain policy belongs in a consuming compliance skill.

Workflow

1. Run the inspector after mapping and change planning:
   • node skills/rye-import-inspector/scripts/inspect_import_run.mjs --source /tmp/source.ndjson --mapped /tmp/mapped.ndjson --change-plan /tmp/change-plan.json --metadata /tmp/import-metadata.json --phase prewrite
2. Fix all blocker findings before target writes.
3. Emit a Rye-stageable report when the inspection result must become part of the run trace:
   • node skills/rye-import-inspector/scripts/inspect_import_run.mjs --mapped /tmp/mapped.ndjson --change-plan /tmp/change-plan.json --metadata /tmp/import-metadata.json --phase prewrite --emit-stage > /tmp/import-inspection-stage.ndjson
4. Include the emitted stage record in the Rye audit NDJSON before tabular_commit_rye.mts.
5. After target writes, run with --phase postwrite and metadata that includes touched IDs plus verification results.

Source-Context Post-Write Gate

For connector/source-context imports that use rye-source-context-intake instead of tabular mapped records, perform an equivalent post-write gate even when the generic inspector script is not the right parser.

Required checks:

• Database counts by node_type.
• Source account/container counts by confirmation_status.
• Source item and artifact counts.
• Candidate counts by status and kind.
• Source container rollup: source items, artifacts, candidates.
• Explicit statement of excluded source types, such as Slack IM/DM/MPIM.
• Confirmation worklist for any source still needs_confirmation.
• Candidate review queue for any candidate still proposed or needs_review.

The run is not ready for accepted knowledge promotion while either condition is true:

• source purpose/default context is unconfirmed and the promotion depends on that source default
• candidate target shape has not been reviewed

Report those as next actions, not as successful completion.

Metadata Contract

Use a small JSON object for process facts that are not already present in mapped records:

{
  "run_id": "recon-aco-update:example",
  "source_reference": "https://...",
  "source_sha1": "abc123",
  "target_tables": ["acquisition_opportunity"],
  "operation_types": ["update"],
  "approval": { "status": "approved", "approved_by": "Casey", "approved_at": "2026-05-16T12:00:00Z" },
  "rye": { "audit_ndjson": "/tmp/recon-rye-audit.ndjson", "committed": true, "run_id": "recon-aco-update:example" },
  "touched_ids": { "acquisition_opportunity": [102238] },
  "verification": { "readback": true, "idempotency": true }
}

Rules

• Local files are inputs only. Rye is the durable traceability record.
• Source rows must preserve source descriptors, raw row fields, and lineage.
• Mapped records must declare destination table, operation, record payload, lineage, and source set.
• Update records must carry old values in mapped metadata or have a change plan with before evidence.
• Target tables and operation types must be declared before production writes.
• Unresolved needs_review or possible_merge findings block writes unless an approval decision is present.
• Post-write checks require touched IDs and readback verification evidence.

The report JSON is the stable interface for downstream compliance skills.