Skip to main content
skill CX Codex

knowledge-base · bis-med-it — Claude Code skill: what it does, how to install

The knowledge-base Claude Code skill helps an agent weigh general econometric knowledge against country-specific data and conventions. It activates when setting up an analysis or working with local sources and research.

Apache-2.0 ★ 89 updated 2026-11-12

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add bis-med-it/marco-BIS --skill knowledge-base -a claude-code
Open on GitHub

What knowledge-base does

knowledge-base sets the order for choosing between general econometric knowledge and information about a particular economy. The agent follows specific guidance in the knowledge base first, falls back to general knowledge when that guidance is missing, and asks the user when uncertain. It also keeps the source of each decision clear.

Users add Markdown documents under knowledge-base/, covering the economy profile, data sources, institutional conventions, literature, and prior beliefs. The repository ships with an empty context/ folder; users create the other materials themselves. Without local material, the skill relies on general knowledge.

What it helps with

  • apply econometric methods in line with a country's conditions
  • choose local data sources instead of generic series
  • account for institutional conventions when selecting a model
  • bring user-provided papers into an analysis context

Example prompts for the agent

  • Set up the analysis using our central bank's data.
  • Use our country's economic profile and modelling conventions.
  • Compare the general method with this paper and our local guidance.

How to install knowledge-base

1Skills CLI (Vercel)

npx skills add bis-med-it/marco-BIS --skill knowledge-base -a claude-code

2Manually: put the files in place

~/.claude/skills/knowledge-base/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/knowledge-base/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about knowledge-base

How do I install knowledge-base in Claude Code?

Run npx skills add bis-med-it/marco-BIS --skill knowledge-base -a claude-code. The skill file is at ~/.claude/skills/knowledge-base/SKILL.md.

What does knowledge-base do?

It tells the agent how to choose between general econometric knowledge and local economic guidance. Specific material in the knowledge base takes precedence over general assumptions.

What can I add to knowledge-base?

Users can add Markdown documents about an economy, local data sources, institutional conventions, relevant literature, and prior beliefs.

Source text

SKILL.md · Apache-2.0
---
name: knowledge-base
description: >
  Manages the interplay between parametric knowledge (the agent's built-in econometric
  training) and specific knowledge (local economic structure, institutional data sources,
  country-specific conventions, user-provided literature). Every other skill in the
  framework consults the knowledge base before making decisions. Use this skill when
  setting up a new analysis context, when the user provides local information or papers,
  or when any skill needs to decide whether to apply general or context-specific logic.
  Also triggers when the user says "use our data", "in my country", "our central bank",
  or provides institutional context.
---

# Knowledge base

## The problem

The agent has two sources of knowledge, and they can conflict:

1. **Parametric knowledge.** What the model learned during training: textbook
   econometrics, canonical papers, standard practices. This is broad but generic,
   and skewed towards US/EU macro.

2. **Specific knowledge.** What the user provides: their country's economic
   structure, institutional data sources, policy regime, modelling conventions,
   relevant local literature. This is narrow but authoritative for the context.

Without explicit orchestration, the agent defaults to parametric knowledge and
may make choices that are technically correct but economically wrong for the
user's context. For example:

- Recommending FRED series when the user's country has its own statistical office.
- Assuming flexible exchange rates when the country has a peg.
- Setting a random-walk prior mean (`rw_deltas` entry of 1) for a variable that is
  stationary in the local economy.
- Choosing a Cholesky ordering based on US institutional timing when the user's
  central bank operates differently.

## Design principle: specific overrides parametric, parametric fills gaps

The full source order (user, dictionary, knowledge base, parametric,
data-driven) is defined in
[`specification/SKILL.md`](../specification/SKILL.md#decision-hierarchy).
Between the knowledge base and parametric knowledge:

1. **If the knowledge base has specific guidance, use it.**
2. **If not, fall back to parametric knowledge.**
3. **If uncertain, ask the user.**
4. **Always make the source of each decision transparent.**

## Knowledge base structure

The knowledge base is a set of documents the user populates, organised by domain.
Each document is a markdown file in the `knowledge-base/` directory.

```
knowledge-base/
├── SKILL.md                 (this file)
├── context/
│   ├── economy-profile.md   country/economy structural description
│   ├── data-sources.md      local data APIs, databases, file locations
│   ├── conventions.md       institutional modelling conventions
│   └── policy-regime.md     monetary/fiscal framework, exchange rate regime
├── literature/
│   ├── paper-summaries/     user-verified summaries of relevant papers
│   └── methodology-notes/   specific methodological guidance from papers
└── priors/
    ├── variable-priors.md   prior beliefs about variable behaviour
    └── identification.md    preferred identification strategies with rationale
```

(Only an empty `context/` folder ships with the repo; its files and the
`literature/` and `priors/` folders are created as the user adds
material.)

### economy-profile.md

The user describes their economy. This informs model selection, variable choice,
and sensibility checks. Template:

```markdown
# Economy profile

## Country/region
[e.g., Euro area, Chile]

## Economic structure
- Open/closed economy: [small open economy / large closed / ...]
- Exchange rate regime: [float / managed float / peg / currency board]
- Capital account: [open / partially open / closed]
- Dominant sectors: [manufacturing, commodities, services, tourism]
- Key trading partners: [...]

## Monetary policy framework
- Instrument: [policy rate / reserve requirements / exchange rate target / ...]
- Target: [inflation targeting / monetary aggregate / exchange rate / dual mandate]
- Transmission channels: [interest rate / credit / exchange rate / asset price]
- Policy meeting frequency: [monthly / 6-weekly / quarterly]
- Decision-to-implementation lag: [immediate / ...]

## Fiscal framework
- Fiscal rule: [yes/no, type]
- Government debt level: [approximate]
- Fiscal year: [calendar / April-March / October-September]

## Known structural features
- Dollarisation: [none / partial / full]
- Informal economy: [estimated share]
- Commodity dependence: [which commodities, share of exports/GDP]
- Seasonal patterns: [agriculture, tourism, fiscal cycles]

## Historical structural breaks
- [Date]: [event and economic significance]
- [Date]: [event]
```

### data-sources.md

Maps conceptual variables to local data sources. This overrides the data-retrieval
skill's default (FRED). Template:

```markdown
# Local data sources

## Primary statistical agency
- Name: [e.g., the national central bank, INE Chile, Eurostat]
- API: [endpoint if available]
- Access method: [API / bulk download / manual / internal database]

## Variable mapping

| Concept           | Local source    | Series ID  | Freq | Notes              |
|-------------------|-----------------|------------|------|--------------------|
| GDP (real)        | [agency]        | [id]       | Q    | [base year, SA]    |
| CPI               | [agency]        | [id]       | M    | [basket details]   |
| Policy rate       | [central bank]  | [id]       | D/M  | [which rate]       |
| Money supply      | [central bank]  | [id]       | M    | [M1/M2/M3]        |
| Exchange rate     | [central bank]  | [id]       | D    | [bilateral/NEER]   |
| Unemployment      | [agency]        | [id]       | M/Q  | [survey details]   |
| ...               | ...             | ...        | ...  | ...                |

## International data
- [Which international sources are also relevant: BIS, IMF IFS, etc.]

## Data quirks
- [e.g., GDP is revised heavily in the first release]
- [e.g., unemployment survey changed methodology in 2012]
- [e.g., CPI basket was rebased in 2019]
```

### conventions.md

Institutional modelling practices. Template:

```markdown
# Modelling conventions

## Preferred software/language
[e.g., MATLAB is standard in our research department / we use R / Python]

## Standard VAR specification
- Typical variable set: [what variables are commonly included]
- Typical lag order: [e.g., 4 for quarterly, 12 for monthly]
- Preferred identification: [e.g., Cholesky, slow variables before the policy rate, fast financial variables after]
- Standard sample period: [e.g., 2000Q1 onwards due to regime change]

## Prior conventions (if Bayesian)
- [e.g., we typically use Minnesota priors with lambda_1 = 0.2]
- [e.g., we treat the exchange rate as I(1) but CPI as trend-stationary]

## Reporting standards
- [e.g., always report 68% and 90% bands]
- [e.g., IRF horizon = 20 quarters]
- [e.g., charts must follow house style: specific colours, fonts]

## Peer review expectations
- [e.g., reviewers expect robustness to lag order and sample period]
- [e.g., sign restrictions are preferred over Cholesky in our department]
```

### policy-regime.md

Describes the policy framework in enough detail for the agent to make sensible
identification and sensibility-check decisions. Template:

```markdown
# Policy regime

## Current regime
- Framework: [e.g., flexible inflation targeting since 2000]
- Instrument: [e.g., 1-day bilateral repurchase rate]
- Target: [e.g., headline CPI inflation 1-3% with 2.5% midpoint]
- Communication: [e.g., forward guidance via policy statement]

## Historical regimes
- [Date range]: [regime description]
- [Date range]: [regime description]

## Implications for modelling
- [e.g., pre-2000 data should not be pooled with post-2000 due to regime change]
- [e.g., the exchange rate was effectively fixed before 1997, so models including
  the exchange rate should start after the float]
- [e.g., unconventional policy measures since 2020 may affect interest rate
  transmission]
```

### literature/ directory

User-provided paper summaries and methodology notes. These are verified by the
user before the agent treats them as authoritative.

Each paper summary should include:
- Citation
- Key contribution (1-2 sentences)
- Model/method used
- Key findings relevant to the user's context
- Any methodological choices the agent should adopt

## How skills consult the knowledge base

Every skill follows this lookup protocol:

### 1. Check for specific guidance first

Before making any decision (variable choice, prior setting, identification,
data source, sensibility threshold), the skill checks the knowledge base:

```
decision_needed: "Which data source for GDP?"

1. Read knowledge-base/context/data-sources.md
   → Found: "GDP (real) from the national statistical office, series <series id>"
   → Use this. Do not default to FRED.

2. If not found in data-sources.md:
   → Fall back to parametric knowledge (FRED, standard series)
   → Inform user: "No local GDP source specified in your knowledge base.
     Defaulting to [FRED/IMF]. You can add your local source to
     knowledge-base/context/data-sources.md."
```

### 2. Check for conflicts

Sometimes parametric knowledge and specific knowledge point in different
directions. Examples:

| Decision | Parametric says | Specific says | Resolution |
|----------|----------------|---------------|------------|
| Variable order | Slow variables, then the policy instrument, then fast variables (e.g. CEE 1999 timing) | Policy rate last (our CB meets after data release) | **Specific wins** |
| GDP prior | Random walk (`rw_deltas` = 1) | Trend-stationary (our GDP is mean-reverting) | **Specific wins** |
| Sample start | 1960Q1 (maximise observations) | 2000Q1 (regime change) | **Specific wins** |
| IRF sign check | Price puzzle is a red flag | Price puzzle expected due to exchange rate pass-through | **Specific wins, but flag for discussion** |
| Lag order | 4 lags for quarterly data | No guidance | **Parametric fills the gap** |

### 3. Make the source transparent

In every decision presented to the user, note whether it came from specific
or parametric knowledge:

> "I'm using the national statistical office's real GDP series [from your data sources].
> For the import price index, I'm defaulting to IMF IFS series
> [parametric; no local source specified]. Would you like to add a local source?"

This transparency lets the user catch errors and progressively fill gaps in
the knowledge base.

### 4. Escalate when uncertain

If the agent encounters a situation where:
- Parametric and specific knowledge clearly conflict, and
- The specific guidance is ambiguous or incomplete

Then ask the user rather than guessing. Frame it as:

> "Your economy profile notes a managed float exchange rate regime, but the
> identification strategy in your conventions assumes a freely floating rate.
> Which should I follow for this analysis?"

## Populating the knowledge base

### Initial setup

The orchestrator's Stage 0a runs the setup, including the fast path that
skips the interview when a run needs no local context (see
[`orchestrator/SKILL.md`](../orchestrator/SKILL.md#stage-0a-knowledge-base-initialisation-one-time)).
Draft each document from the templates above, present it for review, and save
the confirmed version. The knowledge base persists across sessions.

### Incremental updates

As the user works with the framework:

- New papers can be added to `literature/` at any time.
- The agent can suggest updates: "Your analysis used a data source not in your
  knowledge base. Would you like me to add it?"
- After each completed analysis, offer to update conventions based on choices
  made during the session.

### From uploaded papers

When the user provides a paper (PDF or reference):

1. The agent extracts methodological choices relevant to the knowledge base.
2. Presents a summary for user verification.
3. On confirmation, saves it to `literature/paper-summaries/` and updates
   relevant context files if the paper contains local economic information.

This is distinct from the advisor protocol: the advisor handles mathematical
derivation; the knowledge base handles contextual and methodological choices.

## Integration with other skills

| Skill | What it looks up | Knowledge base source |
|-------|-----------------|----------------------|
| Model selection | Economic structure, regime, conventions | economy-profile, policy-regime, conventions |
| Data retrieval | Local data sources, variable mapping | data-sources |
| Specification: BVAR priors | Prior beliefs, variable treatment (I(1) vs I(0)) | variable-priors, conventions |
| Specification: block exogeneity | Foreign block to treat as exogenous (small open economy) | economy-profile |
| Specification: sign restrictions | Expected shock signs in this economy | identification, literature |
| Specification: TVP-VAR | Known structural break dates | economy-profile, policy-regime |
| Specification: local projections | Instruments relevant to this economy | literature, identification |
| Diagnostics | Sensibility benchmarks for this economy | economy-profile, conventions |
| Advisor protocol | Papers to reference for derivations | literature/methodology-notes |
| Research log | Document which knowledge source informed each decision | all |

## Example: a Sims (1980)-style VAR for a small open economy

Without a knowledge base, the agent uses FRED data, the paper's Cholesky
ordering and US macro intuition for the sensibility checks. With one, the same
workflow uses local data (data-sources.md), adds the exchange rate and an
exogenous foreign block (economy-profile.md; see the specification skill's
block-exogeneity step), reorders the Cholesky to local policy timing
(conventions.md), treats the exchange rate as I(1) and the policy rate as I(0)
(variable-priors.md), starts the sample at inflation-targeting adoption
(policy-regime.md), and may expect a price puzzle through exchange rate
pass-through (economy-profile.md, literature/).

© bis-med-it · Apache-2.0 · bis-med-it/marco-BIS

Similar skills

SKILL.md skills catalog →

knowledge-base by other authors comes first.

knowledge-base Inkeep Organizes your repository into a three-layer knowledge base with strict evidence tracking. Keeps raw sources, drafts, and final articles connected. CX Codex +1 875installs firecrawl-knowledge-base firecrawl The firecrawl-knowledge-base skill crawls web pages with Firecrawl and structures them into local Markdown files. It helps prepare clean RAG chunks or fine-tuning datasets. CX Codex +1 33,100installs okf-knowledge-base Inkeep okf-knowledge-base is a Claude Code skill for creating and checking knowledge bundles in Open Knowledge Format. It applies to OKF files, okf plugin warnings, and decisions about types or links. CX Codex +1 9,400installs taiwan-md-knowledge-base reason-machines taiwan-md-knowledge-base — a skill for maintaining an open knowledge base about Taiwan built on Astro v5, with Chinese and English articles and a D3.js graph. Useful if you write or translate content for that project. CX Codex +1 1,300installs knowledge-base-builder OneWave-AI knowledge-base-builder turns support tickets into knowledge base articles: FAQ sections, step-by-step tutorials with screenshot placement, video scripts and deflection metrics. CX Codex +1 688installs cabinet-ai-knowledge-base reason-machines cabinet-ai-knowledge-base is a Claude Code skill for setting up Cabinet, a file-based knowledge base with AI agents and scheduled tasks. It activates for requests to create a workspace or configure Cabinet agents. CX Codex +1 615installs knowledge-base-for-startups aws knowledge-base-for-startups is a Claude Code skill that surfaces AWS Activate program guides, credit rules, and reference architectures. Answers factual questions about startup programs and guides. CX Codex +1 569installs

From the same set12

The whole set →
advisor-protocol The advisor-protocol Claude Code skill sends difficult mathematical reasoning to a more capable model for review. It activates when an agent is unsure how to interpret a derivation or translate it into code. CX Codex +1 ★ 89repo stars econometric-diagnostics econometric-diagnostics is a Claude Code skill that checks VAR/BVAR residuals and model adequacy after estimation or on request to test autocorrelation, heteroskedasticity, normality, or structural breaks. CX Codex +1 ★ 89repo stars model-selection The model-selection Claude Code skill matches econometric models to a research question and available data. It runs when a user needs an approach or wants to check a model they have chosen. CX Codex +1 ★ 89repo stars research-log The research-log Claude Code skill records decisions and stages in an econometric analysis. It runs after another MaRCo@BIS skill finishes a stage or when asked for an audit trail, reproducibility log, or session summary. CX Codex +1 ★ 89repo stars model-dictionary The model-dictionary Claude Code skill searches a literature-based dictionary of econometric specifications. It joins model selection and requests to examine a paper for a new entry. CX Codex +1 ★ 89repo stars benchmarking The benchmarking Claude Code skill evaluates forecast performance and checks estimation results. Run it after a model fit or while preparing results for a paper. CX Codex +1 ★ 89repo stars results-presentation results-presentation — a Claude Code skill that turns econometric results into charts, tables, and written interpretation. It activates at the reporting stage or when results need to be presented or prepared for publication. CX Codex +1 ★ 89repo stars specification This Claude Code skill guides econometric model specification. It activates for choices about variables, lags, priors, identification, or sample periods. CX Codex +1 ★ 89repo stars

Project sponsors

Companies that keep this analytics open