Shared agents, skills, prompts, templates, and workflow tools for OG and CLEWS model work. This repo is a common home for the reusable pieces we build once and want to use across model projects — for example, the OG country calibration skill and shared agent definitions.
| Folder | Contents |
|---|---|
agents/ |
Reusable agent definitions (subagent files, system prompts). |
skills/ |
Skills that package a repeatable task (e.g. OG calibration), one dir per skill. |
prompts/ |
Standalone reusable prompts — analysis, review checklists, briefings. |
templates/ |
Copy-and-fill starting points — config templates, doc skeletons, boilerplate. |
workflows/ |
Multi-step processes that orchestrate agents, skills, and scripts. |
scripts/ |
Small helper scripts (Python, shell) supporting the tools here. |
docs/ |
Longer-form guides, background notes, and reference material. |
Each folder has its own README.md with a bit more detail.
| Tool | Type | Description |
|---|---|---|
| og-country-calibration | skill | Calibrate or refine an OG-Core country model (single- or multi-industry): macro/open-economy parameters, capital share, earnings e-matrix, demographics, tax rates, SAM-based multi-industry splits, and steady-state validation. |
| fable-mode | skill | Fable 5's working discipline as a reusable loop — scope, gather evidence, attack the answer, verify, report. Applies to any model on multi-step, debugging, model-run, or review tasks. |
| clews-model-review | skill | Review a MUIO/OSeMOSYS CLEWs model for structure and data consistency against the NamibiaCLEWs benchmark: referential integrity, orphaned IDs, dangling/stranded technologies, unit discipline, sector coverage, and solve status — via a bundled audit.py checker plus a rubric. |
| add-environmental-accounting | skill | Add a JSON-first environmental accounting layer to any MUIO/OSeMOSYS CLEWS model: inventory physical flows, create separate water and land terminals, preserve existing services, regenerate every scenario through MUIO, and quantify regression differences. |
Keep this repo shareable. Do not commit:
- Secrets — API keys, tokens, passwords, credentials,
.envfiles. - Generated outputs — model runs, solver output, plots, logs, archives.
- Large data — datasets, SAMs, or binaries; link to their source instead.
- Partner-sensitive material — anything that shouldn't be shared broadly.
The .gitignore covers the common cases, but check your git status before
committing.
- One tool per file or directory; give it a short README or header explaining what it is and when to use it.
- Match the existing layout — put things in the folder that fits.
- Keep commit messages to a single line.
- Prefer pointers to large or sensitive data over copies of it.