FAQ

Frequently asked questions and fixes for common issues with InvestSkill. A Traditional Chinese version is on the site: 常見問題.

Installation
Platforms
Choosing and Using Skills
Data and Accuracy
Output and Reports
Versions and Updates
Bugs, Features and Help
Advanced

Installation

Q: What is the quickest way to install InvestSkill?

A: One command, from the project directory where you want to use it. Pick your agent with -a:

curl -fsSL https://raw.githubusercontent.com/yennanliu/InvestSkill/main/install.sh | bash -s -- -a claude
# agents: claude · cursor · copilot · gemini · codex · opencode · any

The installer copies every framework into .investskill/prompts/ and wires up your agent’s own entry point:

`-a`	Wires up
`claude`	`.claude/skills/<skill>/SKILL.md` (slash commands)
`cursor`	`.cursor/rules/investskill.mdc`
`copilot`	`.github/copilot-instructions.md`
`gemini`	`GEMINI.md`
`codex`, `opencode`	`AGENTS.md`
`any`	`.investskill/prompts/` only — paste into any LLM

Run it with -l to print this list, -h for every option.

Q: How do I install it in Claude Code as a plugin?

A: Through the plugin marketplace:

claude
/plugin marketplace add yennanliu/InvestSkill
/plugin install us-stock-analysis@invest-skill
/us-stock-analysis:stock-eval AAPL

The marketplace install gives namespaced commands (/us-stock-analysis:<skill>); install.sh -a claude installs project-level skills you call as /<skill>. Use either — not both in the same project, or every skill will appear twice.

Q: Is it safe to pipe the installer to bash?

A: It only copies markdown files and appends a marked block to your agent’s config file. It never overwrites an existing file — instruction files are appended to between  and  markers — and running it a second time changes nothing. If you’d rather check first, read install.sh or download it and run it locally.

Q: Can I install for my user account instead of one project?

A: Yes — add -g to install into your home directory instead of the current project, or -d DIR to target another directory. -r REF installs from a specific branch or tag.

Q: The plugin doesn’t appear after installing it in Claude Code.

Restart Claude Code (exit, then claude again).
Check it is listed: /plugin list should show us-stock-analysis.
If not, remove and re-add the marketplace: /plugin marketplace remove invest-skill, then /plugin marketplace add yennanliu/InvestSkill and install again.

Q: I get a “plugin not found” error.

A: Check the marketplace is registered with /plugin marketplace list, then install with the full name us-stock-analysis@invest-skill. A network problem during marketplace add is the other common cause — re-run it.

Q: How do I update or uninstall?

A: In Claude Code, /plugin marketplace update invest-skill pulls the latest version; /plugin uninstall us-stock-analysis removes it. For install.sh installs, re-run the same command to update (it refreshes .investskill/prompts/); to uninstall, delete .investskill/ and the InvestSkill:begin … InvestSkill:end block (or the .claude/skills/ / .cursor/rules/investskill.mdc files it created).

Platforms

Q: Which platforms are supported?

A: Claude Code, Cursor, GitHub Copilot, Gemini CLI, Codex, OpenCode, and any chat LLM (ChatGPT, Claude.ai, Gemini web, local models through Ollama). The skills are plain markdown, so anything that can read a prompt can use them. See Platform Compatibility for a side-by-side.

Q: Are all 38 skills available on every platform?

A: Yes. Every platform gets the same 38 skills: 34 analysis frameworks, 3 aliases (fundamental-analysis, dcf-valuation, research-bundle, which redirect to the skill that absorbed them), and the report-generator output tool.

Claude Code: all 38 skills as slash commands
Cursor, Gemini CLI, Codex, OpenCode: all 38 skills as @.investskill/prompts/<skill>.md files (or @prompts/<skill>.md inside a clone of the repo)
Copilot: all 38 skills, referenced from .github/copilot-instructions.md
Any LLM: 38 prompts in prompts/, one per skill

Q: Why is the advertised count 34 and not 38?

A: report-generator is an output tool (it renders other skills’ results as HTML/PDF), not an analysis framework, and the three aliases are kept only so old commands keep working. They ship with the plugin but aren’t counted.

Q: How do I use a skill in Cursor, Gemini CLI, Codex or OpenCode?

A: After install.sh, reference the prompt file in chat:

@.investskill/prompts/stock-eval.md Evaluate AAPL

The rule / instructions file the installer wrote tells the agent the frameworks exist, so you can also just ask (“run the bear-case framework on TSLA”).

Q: Cursor or Gemini doesn’t seem to load the rules.

Check the file exists: .cursor/rules/investskill.mdc (Cursor) or GEMINI.md (Gemini) in the project root.
Make sure you opened / started the agent in that directory — Gemini CLI only reads GEMINI.md from where it was started.
Restart the editor or CLI session.
Ask “What InvestSkill frameworks are available?” — the agent should list them.

Q: @ file references don’t resolve.

A: Check the path: after install.sh the prompts are in .investskill/prompts/, not prompts/. File names use hyphens (stock-eval.md, not stock_eval.md).

Q: How do I use it in ChatGPT, Claude.ai or another chat app?

A: Paste the contents of a prompt file as your first message (or as a custom instruction / project prompt), then ask for the analysis: cat .investskill/prompts/stock-eval.md | pbcopy on macOS.

Q: Can I run it fully offline?

A: Yes, with a local model — see the Ollama guide. The frameworks themselves need no network; only fetching live market data does, and with a local model you paste that data in yourself.

Q: If I start an analysis in Claude Code, can I continue it in Cursor?

A: Not as one session — each platform keeps its own context. The frameworks are identical everywhere, so copy the output across, or use full-report to produce a single HTML file you can share.

Choosing and Using Skills

Q: Which skill should I start with?

A: stock-eval — it covers quality, value and risk in one pass and ends with a go/no-go signal. The Choose a Skill page maps every goal to the right framework.

Q: What happened to fundamental-analysis, dcf-valuation and research-bundle?

A: They were merged into larger skills and are now aliases: fundamental-analysis → stock-eval, dcf-valuation → stock-valuation (DCF is its Method 1), research-bundle → full-report (--depth quick / standard / comprehensive). The old commands still work and point you to the new skill.

Q: Which skill should I use for valuation?

A: stock-valuation — DCF, comparable companies, EV multiples and residual income in one model, with a sensitivity table. Pair it with bear-case to stress the assumptions.

Q: How do I compare several stocks?

A: compare puts 2–5 tickers on one table and argues the pick (/compare AAPL MSFT GOOGL). For a wider list, stock-screener ranks many tickers across valuation, quality, momentum, sentiment and growth.

Q: How do I check whether a company’s earnings or balance sheet can be trusted?

A: forensic-accounting scores manipulation risk (Beneish M-Score, Altman Z, accruals, cash-vs-earnings divergence). credit-check looks at solvency and refinancing risk — the debt maturity wall, interest coverage and liquidity runway.

Q: What should I run around earnings?

A: Before the print: earnings-preview (consensus vs. whisper, implied move, a three-scenario grid). After: earnings-call-analysis on the transcript, then thesis-tracker --update to see whether your thesis still holds.

Q: Does it cover ETFs, taxes and portfolio risk?

A: Yes — etf-analysis (cost, tracking, overlap, structure), tax-lens (US tax mechanics, plus a --non-us module for foreign investors; educational, never tax advice), portfolio-review (allocation and rebalancing) and risk-stress-test (scenario replay, VaR/CVaR, liquidity).

Q: How do I learn from my own trades?

A: Write the thesis with thesis-tracker when you buy, then run trade-postmortem after you sell — it separates process from outcome and pulls out one lesson for your journal. learning-coach explains any skill’s output metric by metric.

Q: Can I use InvestSkill for crypto, forex, commodities or non-US stocks?

A: Not reliably. The frameworks are built around US-listed equities and SEC filings. ETFs, US-listed ADRs and equity options (options-analysis) are covered; crypto, forex and commodities are not.

Q: What if the ticker doesn’t exist or is delisted?

A: The skill will report missing data rather than invent it — every analysis skill has a data-validation gate. Always use the ticker (AAPL), not the company name.

Data and Accuracy

Q: Does InvestSkill fetch live market data?

A: InvestSkill itself fetches nothing — it has no runtime. If your assistant can browse or call tools, the skills tell it which primary sources to read; otherwise it falls back to training data, which is out of date. Every analysis opens with a Data & Sources header so you can see which it used. For current prices, paste them in.

Q: Can I paste my own financial data?

A: Yes, and it is the most accurate way to work. Paste statements, a 10-K / 10-Q / 8-K, an earnings-call transcript or your holdings, then ask for the analysis:

@.investskill/prompts/financial-report-analyst.md
[paste the 10-K text]
Analyze financial quality and red flags

Q: My assistant can’t browse. How do I get the filing?

A: Two optional zero-dependency scripts in the repo fetch the primary source for you to paste: node scripts/fetch-edgar.js AAPL --form 10-K saves the filing as text, and node scripts/fetch-fundamentals.js AAPL builds a reconciled statement pack from the SEC’s XBRL data. They are not part of the plugin. See Data & Accuracy.

Q: How much history should I provide?

A: At least two years of annual reports; five years for trend analysis; ten for cyclical businesses.

Q: How do I check an analysis for hallucinated numbers?

A: Run fact-check on it — it checks every figure against a primary source, recomputes derived numbers and re-issues the report with citations. result-validator then scores data quality, methodology and signal consistency.

Output and Reports

Q: What is the Investment Signal block?

A: Every skill ends with the same box, so results are comparable across skills:

╔══════════════════════════════════════════════╗
║              INVESTMENT SIGNAL               ║
╠══════════════════════════════════════════════╣
║ Signal:      BULLISH / NEUTRAL / BEARISH     ║
║ Confidence:  HIGH / MEDIUM / LOW             ║
║ Horizon:     SHORT / MEDIUM / LONG-TERM      ║
║ Score:       X.X / 10                        ║
╠══════════════════════════════════════════════╣
║ Action:      BUY / HOLD / SELL               ║
║ Conviction:  STRONG / MODERATE / WEAK        ║
╚══════════════════════════════════════════════╝

The Score maps to the Signal (8–10 strongly bullish … 0–1.9 strongly bearish). Confidence should be LOW when the data is thin or from model memory. It is an educational summary, not a recommendation.

Q: Can I get an HTML or PDF report?

A: Yes. full-report AAPL runs several modules and saves a standalone HTML report to output/; report-generator turns any analysis into HTML you can print to PDF. Both work on any platform that can write files; in a chat app, copy the HTML into a file yourself.

Q: Can I get charts?

A: chart-master produces Mermaid, ASCII or Chart.js charts (moving averages, candlesticks, RSI/MACD and more) from the data in your analysis.

Q: Can I get the output in Traditional Chinese?

A: Ask for it in your request. 10k-digest and learning-coach have an explicit --lang zh-TW option.

Versions and Updates

Q: Which version am I on?

A: In Claude Code, /plugin list shows the installed version. In a clone of the repo, jq '.version' plugins/us-stock-analysis/.claude-plugin/plugin.json. The latest release and what changed in each version are in CHANGELOG.md and on the Releases page.

Q: Which versions are supported?

A: The latest minor release. Fixes ship as a new patch release rather than as backports — see SECURITY.md.

Q: How are versions numbered?

A: SemVer. A new skill or capability is a minor release (1.12 → 1.13), fixes and prompt refinements are a patch, and breaking changes to skill names or the output contract are a major release.

Q: Will my analysis change if I upgrade?

A: It can — frameworks are refined between releases, so scores and confidence may shift. The signal block format stays the same within a major version. Re-run anything you rely on after an upgrade, and always verify important decisions independently.

Q: Which Claude Code, Cursor or model version do I need?

A: A recent version of your tool with skill / rules support; there’s no hard minimum beyond that. InvestSkill is plain markdown and runs on any model — more capable models handle multi-module runs like full-report best.

Bugs, Features and Help

Q: How do I report a bug?

A: Open a GitHub Issue with the platform, the version, the exact command, what you expected, and what you got (paste the output in a code block).

Q: How do I suggest a feature?

A: Start a GitHub Discussion with the use case and an example of how you’d use it.

Q: Where do I ask questions?

A: GitHub Discussions for questions, Issues for bugs. The Cookbook has worked examples for every skill.

Advanced

Q: Can I modify a framework?

A: Yes — copy a prompt from prompts/ (or .investskill/prompts/) and edit it. Keep the closing Investment Signal block if you want results to stay comparable.

Q: Can I add my own skill or contribute one?

A: Yes. node scripts/new-skill.js <name> --category … --title … --desc … scaffolds the skill with the full output contract and wires it into every file; see ADDING-NEW-SKILLS.md and CONTRIBUTING.md, then open a PR.

Q: Why is an analysis slow?

A: Multi-module skills (full-report --depth comprehensive) do far more work than a single skill like stock-eval. Use --depth quick or a single skill when you need a fast answer, or switch to a faster model.

Q: Can I use the prompts programmatically?

A: Yes — load a prompt as the system message:

import anthropic

with open("prompts/stock-eval.md") as f:
    framework = f.read()

client = anthropic.Anthropic()
response = client.messages.create(
    model="claude-opus-5",
    max_tokens=16000,
    system=framework,
    messages=[{"role": "user", "content": "Evaluate AAPL"}],
)

Still Need Help?

README — overview and install
Cookbook — examples for every skill
Platform guides — Claude Code, Cursor, Gemini CLI, Ollama
GitHub Discussions · GitHub Issues

Table of Contents