| name | governance-provenance-lead |
| description | Maintains governance, provenance, rights, license, privacy, export-control, safety, and audit evidence across the world-model lifecycle. |
| license | Proprietary |
| metadata | {"owner":"Chris von Csefalvay (HCLTech)","version":"1.1.0","tags":["governance","provenance","rights","audit"]} |
Governance provenance lead
Mission
Use this skill to make the end-to-end world-model capability defensible to engineering, safety, regulatory, legal, and operations leaders. It owns the decision record for its stage of the data-to-world-model-to-sim2real pipeline and must emit artefacts that downstream agents can validate.
Trigger conditions
Activate this skill when the request involves any of the following:
- Governance record and audit evidence map.
- Inputs such as license records, data manifests, model manifests, evaluation decisions, deployment records.
- A handoff into or out of this stage of the autonomy world-model workflow.
- A review of whether the current evidence is sufficient to proceed.
Do not use this skill as a generic brainstorming surface. If required records are missing, create the appropriate manifest skeleton and state the missing evidence rather than inventing values.
Required repository tools
Run these tools from the repository root with PYTHONPATH=src or through scripts/wmb:
| Tool | Use |
|---|
scripts/wmb new-manifest --schema schemas/governance-record.schema.json --output <path> | Create the stage manifest contract. |
scripts/wmb validate-manifest --schema schemas/governance-record.schema.json --manifest <path> | Validate required fields and basic types. |
scripts/wmb readiness-report --root <manifest-dir> --output <report.md> | Summarise evidence completion across manifests. |
scripts/wmb skill-audit --root . --output artifacts/skill-audit.json | Confirm skill packaging depth before release. |
Skill-local helper:
bash skills/governance-provenance-lead/scripts/run-checks.sh <manifest-path>
Required inputs
Before making recommendations, identify and record:
- license records.
- data manifests.
- model manifests.
- evaluation decisions.
- deployment records.
If the user has not provided a value, mark it as missing evidence in the output. Do not create identifiers, customer names, dataset names, platform names, job ids, or benchmark values.
Preflight
- Confirm the request is within this skill's authority and does not require an upstream skill first.
- Locate the relevant manifest, or create a skeleton from
schemas/governance-record.schema.json.
- Run the skill-local check script or
scripts/wmb validate-manifest.
- Classify missing evidence as blocking, non-blocking, or a downstream handoff.
- Check privacy, rights, safety, and deployment authority before recommending any mutating action.
- Name the next owner skill for every unresolved workstream.
Operating workflow
- Map every artefact to owner, permitted use, retention, privacy, export-control, and safety classification.
- Trace source data to dataset versions, training runs, checkpoints, evaluations, services, and incident records.
- Check that agent-generated artefacts are reviewed before use in training, deployment, or safety acceptance.
- Ensure repository content excludes secrets, protected data, gated weights, and operational identifiers.
- Require release records for signatures, scans, benchmarks, approvals, and promotion decisions.
Output contract
Return a concise professional artefact containing:
- governance record.
- artefact lineage map.
- approval gate map.
- retention and access policy.
The output must include:
- Manifest path or expected manifest path.
- Evidence accepted from source records.
- Evidence still missing.
- Human review gates.
- Downstream skill handoffs.
- Commands the operator can run to validate or generate the next artefact.
Quality gates
Do not mark the stage ready unless all applicable gates are satisfied:
- Required manifest fields validate against
schemas/governance-record.schema.json.
- Source records are authoritative and do not rely on invented values.
- Sensitive data and secrets are referenced by controlled handle only.
- Safety and regulatory context are explicitly carried forward.
- Outputs are suitable for review by engineering, safety, legal, quality, and operations owners.
Stop conditions
Stop and request review when any of these conditions appear:
- unlicensed derivative data.
- privacy review absent.
- lineage break.
- approval record missing.
- secret exposure.
Also stop before destructive infrastructure actions, deployment to a real system, model promotion, external publication, or any action involving unapproved customer, medical, regulated, export-controlled, or confidential data.
References to load when needed
references/operating-playbook.md for detailed stage procedure.references/output-contract.md for deliverable structure and evidence tables.BENCHMARK.md for evaluation cases and pass criteria.evals/benchmark-cases.json for machine-readable skill evaluation cases.
