assess-portfolio

name	assess-portfolio
description	Use when the user asks about their portfolio — state, drift, rebalancing, P&L, adding/selling positions, editing targets or cash balance, swapping tickers, or troubleshooting CLI errors. Wraps the `portfolio` CLI of this repo with a structured assessment template and reference docs for every common task.

⚠️ Outdated. The portfolio repo moved to a snapshot-based net-worth model (multiple institutions, ISIN categories, PDF ingest). This skill references the removed transaction/income CLI (add-buy, add-sell, import, income) and needs a rewrite. See docs/superpowers/plans/2026-06-19-multi-source-networth.md.

Assess Portfolio

When to use

Trigger on any of:

• "How's my portfolio doing?" / "Any drift from my targets?" / "Should I rebalance?"
• "Add a buy for X" / "Record that I sold Y" / "I deposited €500"
• "Change my target weight for bonds" / "Add a new category"
• "My ticker shows no price" / "CLI gave me this error"
• Explicit /assess-portfolio invocation

What this skill does NOT do

No open-ended investment advice. This skill does not recommend specific stocks, predict prices, or second-guess the user's strategy. It strictly interprets the user's own data/config.yaml (the targets they set) against portfolio show (their current state) and reports what it finds.

If the user asks for actual financial judgment ("should I buy Nvidia?", "is now a good time to sell bonds?"), decline politely and suggest a licensed financial advisor.

Resources

Scripts (in scripts/)

• snapshot.sh — emits a structured JSON snapshot (positions, totals, rebalance). Prefer this over parsing portfolio show text for agent reasoning. Usage: ./.claude/skills/assess-portfolio/scripts/snapshot.sh
• snapshot.py — the Python source behind snapshot.sh, with the JSON shape documented at the top
• add-transaction.sh — thin wrapper for add-buy / add-sell with action-argument convenience. Usage: ./.claude/skills/assess-portfolio/scripts/add-transaction.sh buy VWCE.DE 10 98.50

References (in references/) — read on-demand based on the user's request

• cli-reference.md — exhaustive docs for every portfolio subcommand, flags, exit codes, examples. Load when the user asks about a specific command.
• data-schema.md — transactions.csv and config.yaml schemas, validation rules, ticker selection guidance. Load when the user is editing data files or picking tickers.
• recipes.md — step-by-step recipes for common tasks (add ticker, rename category, swap delisted ticker, update cash balance, scenario dry-runs). Load when the user wants to make a change.
• troubleshooting.md — error dictionary with meaning + fix for every validation error, yfinance error, and Streamlit issue. Load when the user pastes an error.

Process: assessment requests

For any "how's it going / should I rebalance" question:

Step 1 — Get structured data

Run from the repo root (/Users/huijokim/personal/finance):

./.claude/skills/assess-portfolio/scripts/snapshot.sh

This returns a JSON object with totals, positions, unpriced_tickers, and rebalance — all numbers pre-computed, no text parsing.

If the script fails, cat the stderr and consult references/troubleshooting.md for the fix, then either fix and retry or explain to the user what's broken.

Step 2 — Read the config

cat data/config.yaml to understand the target structure. You need it for context — the snapshot has the numbers but not the user's intent (e.g., why three separate equity categories).

Step 3 — Produce the assessment

Structure the response as four short sections, in this order. 3–5 lines per section.

A. Snapshot (one paragraph)

• Total market value (EUR, including cash)
• Total P&L (EUR + %)
• Cash weight vs. target
• Number of priced vs. unpriced positions (warn if unpriced_tickers is non-empty)

B. Concentration observations

• Largest single position as % of total (flag if >15% and the ticker is not itself a diversified ETF)
• Any category with drift >5pp or less than half its target
• Currency exposure (EUR vs. USD)
• Look-through overlap: if the user holds both a global-equity ETF (VWCE / IWDA) AND a US-equity bucket, note that their effective US weight is higher than the category table shows (MSCI World is ~70% US)

C. Drift vs. target (per category)

Use the rebalance array from the snapshot directly — don't recompute. For each entry:

• |drift_pp| < 1: report as on-target
• 1 ≤ |drift_pp| < 3: note the drift, no action needed yet
• |drift_pp| ≥ 3: recommend the action, quoting the exact delta_eur ("Buy €X" if positive, "Sell €X" if negative)

D. Process suggestions (optional, ≤2 bullets)

Concrete, interpretive observations — not predictions. Examples:

• "IUSA.AS and VUSA.AS are both distributing S&P 500 ETFs — consolidating saves you per-order fees"
• "EUNA.DE has no live price — swap to a ticker yfinance quotes (see recipes.md → swap a delisted ticker)"

Never speculate about market direction, macro, or specific stock picks.

End with one short disclosure: "Numbers above are from a live yfinance fetch at {timestamp} and reflect your config as of this reading."

Process: mutation requests

For "add a buy", "change a target", "rename a category", etc.:

1. Load references/recipes.md and follow the matching recipe.
2. If the user's request doesn't match any recipe exactly, load references/cli-reference.md and/or references/data-schema.md as appropriate.
3. Always run uv run portfolio check after the mutation and report the result.
4. If they asked for a change that affects valuation (new ticker, new weights), consider offering to also run the snapshot and show how the change affected drift.

Process: error troubleshooting

If the user pastes a CLI error:

1. Load references/troubleshooting.md.
2. Match the error text to an entry.
3. Apply the listed fix (either directly, or by guiding the user through the edit).
4. Verify with uv run portfolio check and confirm success.

Output style

• Terse, quantitative, scannable
• Cite concrete EUR amounts from the snapshot, not percentages alone
• Do not re-print data the user has already seen in a tool result
• No per-section disclaimers; one closing line is enough