Use an AI skill to guide a climate-risk assessment, or open its notebook and Python pipeline to inspect and run the analysis yourself. CRC provides open, reproducible hazard and risk workflows. VELO/CDT adds enterprise asset, company and scenario evidence through an authenticated service.
Set the source once per chat, then copy any example from the table:
Use skills from https://github.com/RiskThinking/crc-docs. Before each new
assessment, load the latest skill and its linked files from the same commit.
If you cannot verify the source, tell me before running the assessment.
The skill guides the remaining inputs. Setup covers AI connections and Python installation; no AI tool is required to run the code. Installed skills include this freshness check; older installations need a one-time update. How refresh works.
Skills link to their instructions and include copyable example prompts. You can
also start with just Use the <skill-name> skill. and let the assistant guide
the inputs. Notebooks explain the methods; pipelines run
them headlessly. "Reference" means a related building block, not an end-to-end
implementation of the paired enterprise skill. VELO/CDT skills use live MCP
calls or their linked SDK scripts; there are no enterprise notebook twins here.
| Problem | Skill (CRC) | Skill (VELO/CDT) | Notebook | Pipeline |
|---|---|---|---|---|
| Screen mortgage collateral for river flooding. "What are the 25-, 100- and 500-year depths at these properties?" | Mortgage flood screen Example: Use the crc-screen-mortgage-flood skill for Toronto, Canada. |
Property climate underwriting Example: Use the velo-underwrite-property-climate skill for 392 Markham Street, Toronto. |
JRC flood acquisition; asset evaluation | Acquire flood → screen assets |
| Estimate flood damage under an explicit curve. "How would this approved depth-damage curve affect my insured properties?" | Flood insurance loss Example: Use the crc-model-flood-insurance-loss skill with my attached portfolio and approved depth-damage curve. |
Portfolio insurability triage Example: Use the velo-triage-portfolio-insurability skill for the company in my attached portfolio. |
Event-aligned impacts | Custom curve and assets; fixture example |
| Review physical exposure across an asset portfolio. "Which Frankfurt sites have flood or drought exposure, and where is coverage missing?" | Asset portfolio assessment Example: Use the crc-assess-asset-portfolio-risk skill for Frankfurt, with flood and drought. |
Company or market-index assessment Example: Use the velo-assess-company-climate-risk skill for the S&P 500. |
Asset evaluation; EDO drought | Evaluate separate hazards; fixture evaluation; acquire drought |
| Screen agricultural sourcing areas. "Which sampled corn and soybean areas near Ames overlap modeled flood?" | Agricultural climate risk Example: Use the crc-assess-agricultural-climate-risk skill for corn and soybeans near Ames, Iowa. |
Company assessment when a company/supplier can be resolved; no crop-specific twin Example: Use the velo-assess-company-climate-risk skill for the supplier named in my sourcing assessment. |
Crop and field exposure | Agricultural assessment |
| Map regional flood exposure. "Where do modeled flood cells and candidate places overlap in the Rhine corridor?" | Mortgage flood screen, acquisition reference Example: Use the crc-screen-mortgage-flood skill to guide a regional flood overlay in the Rhine corridor using the flood-by-province notebook. |
- | Flood by province | Flood/admin overlay; covered Overture candidates |
| Understand portfolio tail loss. "What VaR/CVaR follows from these explicit binary outcomes and independence assumptions?" | Asset portfolio assessment, advanced reference Example: Use the crc-assess-asset-portfolio-risk skill to explain the portfolio-risk notebook and its binary-outcome assumptions. |
Company assessment, different model/metrics Example: Use the velo-assess-company-climate-risk skill to review tail-risk metrics for the company in my assessment. |
Portfolio risk metrics | Risk and attribution |
| Explore sensitivity to a changed hazard tail. "What changes if the fitted tail scale rises by 10%?" | Asset portfolio assessment, sensitivity reference Example: Use the crc-assess-asset-portfolio-risk skill to explore a 10% tail-scale stress using the multi-scenario notebook. |
-; local stresses are not enterprise climate projections | Multi-scenario comparison | Local tail stresses |
| Understand a fitted flood curve. "How does a dry-event probability combine with positive flood depth?" | Mortgage flood screen, methods reference Example: Use the crc-screen-mortgage-flood skill to explain the hurdle-distribution primer. |
- | Hurdle-distribution primer | - |
| Reconcile open and enterprise results. "What can each assessment tell us about the same real target?" | Compare CRC and VELO Example: Use the compare-crc-velo-assessments skill with the CRC and VELO results in this conversation. |
Same comparison skill; run the relevant enterprise skill first | - | Artifact inventory, followed by the skill's semantic comparison |
Open a notebook from the table in Colab and choose Runtime → Run all. No clone, API key or GPU is required. The first run installs dependencies and may take several minutes; saved outputs are already visible on GitHub.
Prefer a local environment? Use Python 3.12+:
uv sync --locked
uv run jupyter lab notebooks/For a small fixture-based run from the repository root:
uv run python pipelines/asset_portfolio_pipeline.pyFor live JRC flood acquisition and Overture candidate discovery:
./ai-playbooks/examples/run-open-demo.shThe portfolio learning track uses checked-in Cologne fixtures. The spatial and agricultural examples acquire remote data and may take several minutes. See setup and output files for prerequisites, changing geography and caching.
A completed AI assessment should include a concise results table, relevant charts and a map when locations are available, backed by downloadable data and provenance. The helper scripts produce intermediate data; the skill directs the assistant to assemble the final report. Missing evidence must remain visible.
- Overture places are candidate locations, not proof of ownership, collateral or portfolio membership. Crop cover and predicted field boundaries likewise do not establish yield, ownership or financial exposure.
- Flood depth, damage ratio and portfolio loss are different quantities. A historical JRC result is not a future climate scenario. Keep incompatible hazards and metrics separate.
- Null or unmatched locations are not zero risk. Check source support, units, curve treatment, spatial precision and extrapolation before interpreting values.
- Use approved assets, vulnerability curves and aggregation assumptions for decision-facing work. Illustrative curves and local tail stresses teach methods.
The current reproducible baseline is crc-sdk 0.7.1 / crc-framework 0.2.5, verified against the published packages. See verification results and limits. The SDK handles source access, canonical hazard data and spatial/portfolio workflows; the framework provides distributions, fitting, impacts and risk metrics.
- Capability matrix: supported surfaces and interpretation limits.
- Gap backlog: remaining workflow work, distinct from available primitives.
- Product strategy: why pair open and enterprise evidence.
- Agent entry point: canonical skill discovery.
CRC and VELO/CDT results require an explicit semantic crosswalk; this repository does not claim that all enterprise computations already run on CRC.