Claw-Screener Stock Screener
Combines Williams %R oversold signals with Buffett-style fundamental analysis to surface quality oversold stocks in the US and Thai markets.
Evidence shows local read-only analysis scripts writing only self-declared watchlist and cache files; no credential access, no covert exfiltration, no destructive defaults. Data flows (SEC EDGAR, Yahoo Finance, Treasury) are disclosed in SKILL.md and README, with a dedicated file-persistence section. Deducted for: no user-confirmation gates, manual-only cache/watchlist cleanup, no rollback mechanism, unverified publisher identity.
Docs and scripts are highly self-consistent (npm scripts, option tables, exit codes, Pitfalls, fallbacks like DCF defaulting to 10%); repo includes tests (config/database/formulas) and a CI workflow. Deducted for: static review cannot execute reproduction; key screening/watchlist paths (screening, watchList, technicalIndicators) lack test coverage, and network-failure feedback for Yahoo/SEC rests on documentation only.
Trigger description is clear (oversold screening, Buffett formulas, Carlson compounders, single-stock analysis); non-fit boundaries declared (Thai market has no SEC fundamentals; bk is a th alias), with quick-test guidance. Deducted for: no Chinese-language support; core function depends entirely on Yahoo Finance, SEC EDGAR and Treasury services that are unreachable or unstable from mainland-China networks, with no mirror or alternative source.
SKILL.md is well layered (install, tools, examples, pitfalls, caching, config, verification); MIT license, version 1.0.0, author and repository stated. Deducted for: no CHANGELOG, no explicit maintenance/update commitment, SEC_USER_AGENT placeholder requires user setup, unverified publisher.
Core tasks (screening, analysis, watchlist) have complete commands, parameters, output formats (text//telegram) and examples; marginal value over manual screening is plausible. Deducted for: static review cannot verify output correctness; first full-universe run costs 20-30 minutes; the investment efficacy of DCF/scoring formulas is author-asserted only.
Auditable primary material exists: test files, CI config, package-lock dependency pinning, detailed docs and verification commands. Deducted for: CI run results not visible in provided files, tests do not cover all key paths, no independent third-party reproduction, output examples are author-authored demos.
- This score is a static source review only; no scripts were executed and actual runtime behavior is unverified.
- Core functionality depends on Yahoo Finance, SEC EDGAR and US Treasury services that may be unreachable or unstable from mainland-China networks, with no fallback data source.
- Full-universe first scans take ~20-30 minutes; validate with --max-tickers or specific --tickers first.
- SEC requires a real contact-email User-Agent; set the SEC_USER_AGENT env var yourself.
- Screening results are quantitative indicators only, not investment advice; formula efficacy is independently unverified.
- Publisher is not verified by the FollowSkills registry; identity is treated as unknown.
What does this skill do, and when should you use it?
Claw-Screener is a Node.js stock screening skill that finds technically oversold stocks using the Williams %R indicator, then scores them against Warren Buffett's ten formulas using SEC EDGAR data. It covers US (S&P 500) and Thai (SET) markets, adds a Carlson-style compounder screen (ROIC, buybacks, margins), and includes a watchlist with deduplicated alerts. Output can be plain text, JSON, or Telegram format. The skill ships as a full repository with TypeScript scripts and SQLite caching.
Fetches prices from Yahoo Finance and US fundamentals from SEC EDGAR, then runs five commands: combined screening (Williams %R < -80 plus a Buffett score threshold), a technical-only oversold scan, deep single-stock analysis via Buffett's 10 formulas, a Compounding Machine scan (revenue/net income trends, ROIC > 15%, 3-year share reduction, operating margin > 20%, with yield-vs-5y and a simple 10-year DCF), and watchlist add/remove/update/list/status/check with oversold, overbought, and quality alerts. Results output as text, JSON, or Telegram messages; fundamentals and prices are cached in SQLite files (default TTLs of 7 and 1 day), and the watchlist persists at ~/.claw-screener-watchlist..
- A value investor hunting buy-the-dip candidates can run combined screening with --min-score 7 to find oversold stocks with strong fundamentals
- A long-term US stock holder wanting to check whether a holding passes Buffett's criteria can run deep analysis on a single ticker
- A Thai SET market follower can run the technical-only scan (--market bk) for oversold signals, since SEC fundamentals are unavailable there
- A quality-compounder investor can use the Compounding Machine to filter for ROIC, buyback, and margin thresholds across the S&P 500
- A trader wanting routine alerts can maintain a watchlist and run watchlist:check on a schedule for oversold/overbought/quality notifications
What are this skill's strengths and limitations?
- Screens on both technical and fundamental dimensions rather than a single indicator
- Covers two markets (US and Thailand), with a fundamentals-free technical scan usable for Thai tickers
- SQLite caching (tunable TTL) makes repeat runs much faster
- Three output formats (text/JSON/Telegram) suit automation and messaging workflows
- Includes DCF valuation context driven by the live 10-year Treasury yield
- First uncached full-US-universe Compounding Machine run takes roughly 20-30 minutes
- Thai market gets price data only — no fundamental analysis
- Depends on Yahoo Finance, whose rate limits can slow bulk scans
- No backtesting or performance validation is documented; screening output is not investment advice
- Requires Node.js >=20 and must be run from the repo root or commands fail
How do you install this skill?
Requires Node.js >=20, npm, and network access to SEC EDGAR and Yahoo Finance. For Hermes: git clone https://github.com/rsoutar/claw-screener.git ~/.hermes/skills/finance/claw-screener, then cd into it and run npm install; alternatively point skills.external_dirs in ~/.hermes/config.yaml at an existing clone. For OpenClaw/Clawdbot: run openclaw skills install git:rsoutar/claw-screener@main, then cd {baseDir} && npm install. All commands must be run from the repository root.
How do you use this skill?
Run npm scripts from the repo root, placing CLI flags after --. Examples: npm run screening -- --market us --min-score 7 --top-n 5 for combined screening; npm run technical -- --market bk for a Thai technical scan; npm run analyze -- AAPL for single-stock analysis; npm run compounder -- --tickers AAPL,MSFT,NVDA for a quick compounder run; npm run watchlist:add -- AAPL,MSFT to track stocks. If SEC EDGAR shows the placeholder User-Agent warning, export SEC_USER_AGENT="YourName/1.0 ([email protected])". See SKILL.md for the full flag reference.