gemelnet-advisor

name

gemelnet-advisor

description

Query and analyse Israeli provident (קופת גמל), study (קרן השתלמות) and investment-provident (גמל להשקעה) and pension funds using the official GemelNet open dataset on data.gov.il. Search funds, show a single fund snapshot, compare funds, rank one against its like-for-like peers, scan a statement into a portfolio checkup, and estimate the shekel cost of management fees. Public fund-level data only — framed as data-grounded analysis, not financial advice. Ships a UserPromptSubmit hook that auto-triggers on Hebrew/English fund keywords.

GemelNet Advisor — Israeli Pension Funds

Analyse Israeli long-term-savings funds from the regulator's own public data: the GemelNet open dataset published by the Ministry of Finance (Capital Market, Insurance & Savings Authority) on data.gov.il. The engine pulls live, fund-level, net-of-fee figures — returns, fees, assets — and turns them into side-by-side comparisons and peer rankings.

Not financial advice. Everything here is historical, public, fund-level data presented as analysis. It contains no personal data and no secrets. Tell the user this explicitly; do not make buy/sell/switch recommendations or promises about future returns.

---

When to use

Trigger this skill when the user asks about — in Hebrew or English:

• A specific Israeli fund (provident / study / investment-provident / pension).
• "Compare these funds", "which has better returns / lower fees".
• "Is my fund any good?" / "rank my fund against its peers".
• "How much am I paying in fees?" (shekel cost of דמי ניהול).
• A pension statement / portfolio checkup.

The bundled UserPromptSubmit hook (hooks/gemelnet-hook.py) auto-suggests the skill when it sees keywords like גמל, קרן השתלמות, פנסיה, תשואה, gemel, provident fund, study fund, pension.

---

The engine

scripts/gemelnet.py — Python standard library only, no pip install. It talks to the data.gov.il CKAN datastore API. Run from the skill folder:

# List the GemelNet dataset's resource IDs (sanity check — needs network)
python3 scripts/gemelnet.py resources

# Search funds (Hebrew or English free text)
python3 scripts/gemelnet.py funds --q "אלטשולר"
python3 scripts/gemelnet.py funds --q "study" --limit 15

# One fund's latest snapshot
python3 scripts/gemelnet.py fund 9012

# Compare funds side by side
python3 scripts/gemelnet.py compare 9012 512 1234

# Rank a fund against its like-for-like peers (same classification)
python3 scripts/gemelnet.py rank 9012

# Estimate the annual shekel cost of the management fee on a balance
python3 scripts/gemelnet.py revenue 9012 --balance 250000

Each command prints a compact table and ends with a "not advice" reminder. See references/api.md for the dataset/field details and references/analysis.md for how to read returns, fees and peer rankings responsibly.

---

Workflow

1. Identify the fund(s). If the user gives a fund number, use it. Otherwise run funds --q "<name>" (Hebrew accepted) and confirm the right FUND_ID before going further. Fund numbers are the stable key; names repeat.
2. Pull the data. Use fund, compare, or rank as the question demands. Always compare within the same classification (FUND_CLASSIFICATION) — a general provident fund and a bond-only study fund are not peers.
3. Read it responsibly (see references/analysis.md):
   • Prefer 3yr / 5yr average-annual returns over a single month or YTD.
   • State returns net of fees, and surface the management fee (דמי ניהול) separately — a small fee gap compounds.
   • For "how much do I pay", use revenue --balance N; explain it's a flat-balance estimate that ignores deposits, deposit fees and compounding.
4. Statement / portfolio checkup ("scan"). When the user pastes or uploads a statement, extract each fund number + balance, then for each holding run fund (snapshot), rank (peer standing) and revenue (fee cost). If a PDF is provided, try pdfplumber / pypdf / pdftotext to pull the text; if none is installed, ask the user to paste the fund numbers and balances. Roll the per-fund results into one checkup table.
5. Frame the answer. Lead with the data, name the report period, and close with the standing disclaimer: public, historical, fund-level data — analysis, not financial advice.

---

Guardrails

• Public data only. Fund-level figures from GemelNet — never ask for, store, or echo back national-ID numbers, account credentials, or other PII.
• No advice. No buy/sell/switch calls, no future-return promises, no tax or legal guidance. Offer comparisons and context; let the user decide.
• Be precise about provenance. Always name the source (GemelNet / data.gov.il) and the report period. Past performance ≠ future results.
• Degrade gracefully. If the network or a resource is unavailable, say so and show the resources output rather than guessing numbers.

---

Files

gemelnet-advisor/
├── SKILL.md                  this file — the workflow
├── README.md                 overview + install + usage
├── skill.json                catalog metadata
├── .gitignore                ignores __pycache__, *.pyc, local scans
├── scripts/
│   └── gemelnet.py           the engine (stdlib only)
├── references/
│   ├── api.md                GemelNet dataset + CKAN API + fields
│   └── analysis.md           how to read returns / fees / rankings
└── hooks/
    └── gemelnet-hook.py      UserPromptSubmit auto-trigger