Personal Finance Skill (AI-Powered Personal CFO)
Unify banking, brokerage, tax, market intelligence, and social sentiment into one skill of 75 tools that replaces AI-guessed numbers with deterministic calculations.
Positives: SKILL.md mandates finance_policy_check before side effects, explicit live-trade confirmation, no credential exposure, paper-env default, clear risk tiers. Deductions: policy engine implementation not present in evidence — only documented promise; handles highly sensitive banking/brokerage/tax data with no encryption, retention or rollback disclosure; onboard.sh requires running an unaudited script; publisher unverified.
Positives: repo contains real vitest suites for alpaca-trading and finance-core covering order confirmation, limits, idempotent storage and normalization, with specific error messages. Deductions: static review cannot execute; tax-engine (23 tools, largest extension) has no test evidence; the '612 passing' claim is unverifiable; failure feedback only partially visible.
Positives: explicit trigger scenarios and 9 workflows, clear tool catalog, extensions usable independently. Deductions: no declared non-fit boundaries; all core function depends on overseas services (Plaid, Alpaca, IBKR, Finnhub, X) with no Chinese-language support — a substantive mainland-China reachability barrier, disclosed and deducted here.
Positives: MIT license, version 2.0.0, layered documentation (SKILL.md → references/ → api-*.md), onboarding script, developer guide. Deductions: no CHANGELOG, no named maintainer or update path, author listed only as 'contributors', incomplete versioning governance.
Positives: clear design goals (deterministic math, policy gating, snapshot storage), workflows map directly to user tasks, marginal value over manual tracking plausible. Deductions: static review cannot verify output correctness or direct usability; tax-calculation accuracy (a high-stakes domain) lacks independent evidence; cost/benefit unargued given multiple paid API credentials.
Positives: auditable test source code as primary material, README states test counts and run commands. Deductions: no visible CI workflow, passing status not independently reproducible, coverage claims (55+ files) unverifiable, marketing claims (40+ agent compatibility) unsupported.
- This is a static source-only review; no code was executed, and confidence is low.
- The policy engine and approval gating are documented but their implementation was not present in the evidence; live-trading risk must be verified by the user.
- Core function depends entirely on overseas services (Plaid, Alpaca, IBKR, Finnhub, X); mainland-China network reachability is a substantive barrier and there is no Chinese-language support.
- Handles highly sensitive banking, brokerage and tax data; encryption, retention and rollback mechanisms are undisclosed.
- onboard.sh builds and registers plugins on the local machine; audit the script before running.
- Tax-calculation tools (liability, TLH, AMT, etc.) are unverified for accuracy; consult a qualified advisor for tax decisions.
- No CHANGELOG or named maintainer; long-term maintenance status unknown.
What does this skill do, and when should you use it?
This is a personal finance management skill for the Agent Skills Protocol: 7 extensions, 75 tools, version 2.0.0, MIT-licensed. It aggregates bank accounts via Plaid, executes trades via Alpaca, monitors portfolios via IBKR, and ships a tax engine with 15 IRS form parsers and 8 calculators. All money math is done by deterministic tools rather than LLM arithmetic, and every trade or tax move must pass a policy check and approval gates. The repo claims 612 passing tests, but it depends on many third-party API keys, so setup effort and running costs are nontrivial.
Calls the Plaid API to pull balances, transactions, investments, and liabilities, then stores normalized snapshots; calls Alpaca to place/cancel orders and read positions and market data (live orders require human confirmation); reads positions, allocation, and NAV history through the IBKR Client Portal; runs a tax engine that parses W-2, 1099-B/DIV/INT, K-1, Form 1040, Schedules A-E/SE, 8949, and 6251, estimates liability, finds tax-loss-harvesting candidates, checks wash-sale compliance, compares FIFO/LIFO/specific-ID lots, and estimates quarterly payments, plus state tax for CA/NY/NJ/IL/PA/MA/TX/FL and AMT; fetches news, fundamentals, analyst ratings, and macro series from Finnhub, SEC EDGAR, FRED, BLS, and Alpha Vantage; pulls sentiment from StockTwits and X and congressional trading data from Quiver; performs anomaly detection (unusual transactions, balance drops, duplicate charges), subscription tracking, net worth and cash-flow summaries, and supports openclaw cron-based scans and weekly briefings.
- An individual with multiple bank and brokerage accounts who wants consolidated net worth and monthly cash flow via natural language
- A US stock investor looking for year-end tax-loss harvesting opportunities with automatic wash-sale checks and lot selection
- A taxpayer holding W-2s, 1099s, and K-1s who wants forms parsed and quarterly estimated taxes computed
- A multi-broker investor (Alpaca + IBKR) who wants allocation-drift monitoring with alerts when positions stray from target bands
- A retail researcher combining SEC filings, company news, StockTwits sentiment, and congressional trading disclosures for single-name analysis
- Anyone who wants a daily automated scan for unusual transactions, duplicate charges, and balance drops
What are this skill's strengths and limitations?
- Unusually broad coverage: 75 tools spanning banking, trading, portfolio, tax, market intel, and social sentiment on one canonical data model
- Enforced deterministic math — tax amounts, P/L, and net worth are never LLM arithmetic, reducing hallucinated numbers
- Built-in policy engine and approval gates; live trades always require explicit human confirmation, backed by 8 hard guardrails
- Deep tax engine: 15 form parsers plus 8 calculators including AMT, 8 state tax computations, and wash-sale validation
- Repo claims 612 passing tests with extensive references/ documentation
- Requires up to 9 third-party API keys, many with paid tiers — real running costs
- Cron scheduling and plugin registration are tied to the OpenClaw runtime; other users must adapt cron themselves
- IBKR depends on a local Client Portal Gateway needing a ~1-minute heartbeat (ibkr_tickle) — noticeable operational burden
- Trading only supports Alpaca and IBKR; no other brokers, and state tax covers only 8 states
- The 612-test claim is self-reported in the README with no independent verification; docs explicitly state outputs are not financial advice
How do you install this skill?
Clone the repo and run the one-command onboard script: git clone https://github.com/6missedcalls/personal-finance-skill.git && cd personal-finance-skill && ./scripts/onboard.sh. The script installs dependencies, builds all 7 extensions, registers each as an OpenClaw plugin (openclaw plugins install -l), and symlinks the skill into ~/.openclaw/skills/. Alternatively install via npx skills add <owner>/personal-finance-skill (with -a claude-code -a cursor -a codex to target specific agents). Then export only the API keys you need (PLAID_CLIENT_ID/SECRET, ALPACA_API_KEY/SECRET, IBKR_BASE_URL, FINNHUB_API_KEY, FRED_API_KEY, etc.). Installation details for agents other than OpenClaw are not fully documented in the source.
How do you use this skill?
Trigger it with natural language, e.g. "compute my net worth and this month's savings rate", "find tax-loss harvesting candidates in my portfolio and check wash-sale risk", or "parse my 1099-B and estimate this year's tax liability". For automation, e.g. a weekly brief: openclaw cron add --name "Finance Weekly Brief" --cron "0 8 * * 1" --message "Sync all providers, compute net worth delta, generate a brief". Every order, transfer, or tax move passes finance_policy_check first; Alpaca can run in paper mode (ALPACA_ENV=paper) for rehearsal.
How does this skill compare with similar options?
The README positions this skill against "most AI finance tools that hallucinate numbers", emphasizing deterministic math and mandatory policy checks. It also claims compatibility with 40+ AI agents (Claude Code, Cursor, Codex, Windsurf, etc.) via the Agent Skills Protocol — its differentiation versus single-platform finance plugins. The source names no other specific competitors.