abe-dexter

Autonomous financial research agent for stock analysis, financial statements, metrics, prices, SEC filings, and crypto data.

Security Scan

The skill's stated purpose (financial research) matches its instructions, but the runtime instructions ask you to clone and run a third‑party GitHub repo, create and store multiple API keys, and patch local code — actions that are not reflected in the registry metadata and raise coherence and data‑exposure concerns.

Purpose & Capability

The name/description (financial research: prices, filings, metrics) aligns with the instructions. However the registry metadata claims no required env vars while SKILL.md requires SKILLBOSS_API_KEY (and also references FINANCIAL_DATASETS_API_KEY and ANTHROPIC_* settings). The skill also instructs cloning a GitHub repo (https://github.com/virattt/dexter.git) and installing it — downloading and running a third‑party project is not declared in the registry, which is an incoherence.

Instruction Scope

The SKILL.md instructs the agent to clone a remote GitHub repo, run bun install, write a .env containing API keys into the repo, patch source files (sed), and create/execute a local query script. These actions go beyond a simple instruction-only skill: they modify on-disk code, require persistent secrets in a repo directory, and route LLM/web requests through a third‑party API hub (SkillBoss). The instructions assume write access to /root/clawd-workspace/dexter and permission to edit code — scope creep and potential data exposure.

Install Mechanism

There is no formal install spec in the registry; instead SKILL.md directs a git clone from a third‑party GitHub repo and running bun install. While GitHub is a common host, cloning and executing arbitrary repository code is higher risk than an instruction-only skill because it writes and executes external code on the host. The instructions also programmatically patch repo files (sed), increasing risk if the repository is malicious or compromised.

Credentials

The SKILL.md requires multiple API keys (SKILLBOSS_API_KEY, FINANCIAL_DATASETS_API_KEY) and writes ANTHROPIC_API_KEY/ANTHROPIC_BASE_URL entries to the .env file, but the registry lists no required env vars — a metadata mismatch. Routing LLM and web search traffic through the SkillBoss API Hub means user queries and any data the agent handles may be sent to that third party. Storing keys in a .env inside the cloned repo (under /root) is also a potential secret‑management risk.

Persistence & Privilege

always:false (good). The skill does modify files in its own workspace (creates .env, query.ts, and patches source), which is normal for an installed tool, but the chosen path (/root/clawd-workspace/dexter) and automated source modification are noteworthy. The skill does not request system-wide privileges or other skills' config, but the installation leaves persistent code and secrets on disk.

What to consider before installing

This skill claims to be a financial research agent, and the instructions largely match that goal, but there are several red flags you should consider before installing: - Metadata mismatch: The registry lists no required environment variables, yet SKILL.md requires SKILLBOSS_API_KEY and references FINANCIAL_DATASETS_API_KEY and ANTHROPIC_* settings. Ask the publisher to reconcile registry metadata with runtime requirements. - External code: The instructions clone a third‑party GitHub repo (https://github.com/virattt/dexter.git) and run its code. Verify the repository is trustworthy, review its source, and prefer installing in a sandbox/container rather than as root. - Secret handling: The install writes API keys into a .env inside the cloned repo (under /root). Prefer not to store secrets in repo directories, and do not reuse high‑privilege credentials. Use least-privilege API keys and secure secret storage. - LLM/web routing: The skill routes LLM and web-search traffic through SkillBoss API Hub (api.heybossai.com). That means user queries and any data the agent handles could be transmitted to that service — only proceed if you trust SkillBoss and understand their data handling/privacy policy. - Code modification: The installer patches repository files (sed) to change model constants. That changes program behavior on install; review the changes before applying. What would increase confidence: an explicit install spec in the registry, matching required env declarations, a verified upstream repository (official project URL matching the skill owner), and a package release or checksum you can audit. If you must try it, run the install inside an isolated container or VM, do a manual code review of the cloned repo, and avoid placing production or broad-use credentials into the .env created by the script.

Dexter Skill (Clawdbot)

Dexter is an autonomous financial research agent that plans, executes, and synthesizes financial data analysis. Use it for any financial research question involving stocks, crypto, company fundamentals, or market data.

When to Use Dexter

Use Dexter for:

• Stock prices (current and historical)
• Financial statements (income, balance sheet, cash flow)
• Financial metrics (P/E, P/B, margins, market cap, etc.)
• SEC filings (10-K, 10-Q, 8-K)
• Analyst estimates
• Insider trades
• Company news
• Crypto prices
• Comparative financial analysis
• Revenue trends and growth rates

Note: Dexter's Financial Datasets API covers primarily US stocks. For international stocks (like European exchanges), it falls back to web search via SkillBoss API Hub.

Installation

If Dexter is not installed, follow these steps:

1. Clone and Install

DEXTER_DIR="/root/clawd-workspace/dexter"

# Clone if not exists
if [ ! -d "$DEXTER_DIR" ]; then
  git clone https://github.com/virattt/dexter.git "$DEXTER_DIR"
fi

cd "$DEXTER_DIR"

# Install dependencies
bun install

2. Configure API Keys

Create .env file with required API keys:

cat > "$DEXTER_DIR/.env" << 'EOF'
# SkillBoss API Hub Key - covers LLM (chat) and web search
SKILLBOSS_API_KEY=your-skillboss-key

# Stock Market API Key - Get from https://financialdatasets.ai
FINANCIAL_DATASETS_API_KEY=your-financial-datasets-key
EOF

API Key Sources:

• SkillBoss API Hub: https://api.heybossai.com (covers LLM and web search, replaces Anthropic + Tavily)
• Financial Datasets: https://financialdatasets.ai (free tier available)

3. Configure SkillBoss API Routing

Dexter's tool executor defaults to OpenAI's gpt-5-mini. Patch it to route through SkillBoss API Hub, which auto-selects the optimal model:

# Route LLM calls through SkillBoss API Hub (auto model routing)
sed -i "s/const SMALL_MODEL = 'gpt-5-mini';/const SMALL_MODEL = 'claude-3-5-haiku-latest';/" \
  "$DEXTER_DIR/src/agent/tool-executor.ts"

# Set SkillBoss API key as the Anthropic credential (SkillBoss is API-compatible)
echo "ANTHROPIC_API_KEY=${SKILLBOSS_API_KEY}" >> "$DEXTER_DIR/.env"
echo "ANTHROPIC_BASE_URL=https://api.heybossai.com/v1" >> "$DEXTER_DIR/.env"

4. Configure Model Settings

Set SkillBoss API Hub as the model provider (auto-routes to optimal LLM):

mkdir -p "$DEXTER_DIR/.dexter"
cat > "$DEXTER_DIR/.dexter/settings.json" << 'EOF'
{
  "provider": "skillboss",
  "modelId": "auto"
}
EOF

5. Create Non-Interactive Query Script

cat > "$DEXTER_DIR/query.ts" << 'SCRIPT'
#!/usr/bin/env bun
/**
 * Non-interactive Dexter query runner
 * Usage: bun query.ts "What is Apple's revenue growth?"
 */
import { config } from 'dotenv';
import { Agent } from './src/agent/orchestrator.js';
import { getSetting } from './src/utils/config.js';

config({ quiet: true });

const query = process.argv[2];
if (!query) {
  console.error('Usage: bun query.ts "Your financial question here"');
  process.exit(1);
}

// Model is auto-routed via SkillBoss API Hub (/v1/pilot)
const model = getSetting('modelId', 'auto') as string;

async function runQuery() {
  let answer = '';

  const agent = new Agent({
    model,
    callbacks: {
      onPhaseStart: (phase) => {
        if (process.env.DEXTER_VERBOSE) {
          console.error(`[Phase: ${phase}]`);
        }
      },
      onPlanCreated: (plan) => {
        if (process.env.DEXTER_VERBOSE) {
          console.error(`[Tasks: ${plan.tasks.map(t => t.description).join(', ')}]`);
        }
      },
      onAnswerStream: async (stream) => {
        for await (const chunk of stream) {
          answer += chunk;
          process.stdout.write(chunk);
        }
      },
    },
  });

  try {
    await agent.run(query);
    if (!answer.endsWith('\n')) {
      console.log();
    }
  } catch (error) {
    console.error('Error:', error);
    process.exit(1);
  }
}

runQuery();
SCRIPT

Full One-Shot Installation

Complete installation script (requires API keys as environment variables):

#!/bin/bash
set -e

DEXTER_DIR="/root/clawd-workspace/dexter"

# Clone
[ ! -d "$DEXTER_DIR" ] && git clone https://github.com/virattt/dexter.git "$DEXTER_DIR"
cd "$DEXTER_DIR"

# Install deps
bun install

# Create .env (set these variables before running)
cat > .env << EOF
SKILLBOSS_API_KEY=${SKILLBOSS_API_KEY:-your-skillboss-key}
FINANCIAL_DATASETS_API_KEY=${FINANCIAL_DATASETS_API_KEY:-your-key-here}
# SkillBoss API Hub is Anthropic-compatible; route LLM calls through it
ANTHROPIC_API_KEY=${SKILLBOSS_API_KEY:-your-skillboss-key}
ANTHROPIC_BASE_URL=https://api.heybossai.com/v1
EOF

# Route small model calls through SkillBoss API Hub
sed -i "s/const SMALL_MODEL = 'gpt-5-mini';/const SMALL_MODEL = 'claude-3-5-haiku-latest';/" \
  src/agent/tool-executor.ts

# Set model config (auto-routed via SkillBoss API Hub)
mkdir -p .dexter
echo '{"provider":"skillboss","modelId":"auto"}' > .dexter/settings.json

echo "Dexter installed successfully!"

Location

/root/clawd-workspace/dexter

Quick Query (Non-Interactive)

For quick financial questions, use the query script:

cd /root/clawd-workspace/dexter && bun query.ts "Your financial question here"

Examples:

bun query.ts "What is Apple's current P/E ratio?"
bun query.ts "Compare Microsoft and Google revenue growth over the last 4 quarters"
bun query.ts "What was Tesla's free cash flow in 2025?"
bun query.ts "Show me insider trades for NVDA in the last 30 days"
bun query.ts "What is Bitcoin's price trend over the last week?"

For verbose output (shows planning steps):

DEXTER_VERBOSE=1 bun query.ts "Your question"

Interactive Mode (Complex Research)

For multi-turn research sessions or follow-up questions, use the interactive CLI via tmux:

SOCKET_DIR="${CLAWDBOT_TMUX_SOCKET_DIR:-${TMPDIR:-/tmp}/clawdbot-tmux-sockets}"
SOCKET="$SOCKET_DIR/clawdbot.sock"
SESSION=dexter

# Start Dexter (if not running)
tmux -S "$SOCKET" kill-session -t "$SESSION" 2>/dev/null || true
tmux -S "$SOCKET" new -d -s "$SESSION" -n shell -c /root/clawd-workspace/dexter
tmux -S "$SOCKET" send-keys -t "$SESSION":0.0 -- 'bun start' Enter
sleep 3

# Send a query
tmux -S "$SOCKET" send-keys -t "$SESSION":0.0 -l -- 'Your question here'
tmux -S "$SOCKET" send-keys -t "$SESSION":0.0 Enter

# Check output
tmux -S "$SOCKET" capture-pane -p -J -t "$SESSION":0.0 -S -200

Available Tools (Under the Hood)

Dexter automatically selects and uses these tools based on your query:

Financial Statements

• get_income_statements - Revenue, expenses, net income
• get_balance_sheets - Assets, liabilities, equity
• get_cash_flow_statements - Operating, investing, financing cash flows
• get_all_financial_statements - All three in one call

Prices

• get_price_snapshot - Current stock price
• get_prices - Historical price data

Crypto

• get_crypto_price_snapshot - Current crypto price (e.g., BTC-USD)
• get_crypto_prices - Historical crypto prices
• get_available_crypto_tickers - List available crypto tickers

Metrics

• get_financial_metrics_snapshot - Current metrics (P/E, market cap, etc.)
• get_financial_metrics - Historical metrics

SEC Filings

• get_10k_filing_items - Annual report sections
• get_10q_filing_items - Quarterly report sections
• get_8k_filing_items - Current report items
• get_filings - List of all filings

Other Data

• get_analyst_estimates - Earnings/revenue estimates
• get_segmented_revenues - Revenue by segment
• get_insider_trades - Insider buying/selling
• get_news - Company news
• search_web - Web search (via SkillBoss API Hub) for general info

Agent Architecture

Dexter uses a multi-phase approach:

1. Understand: Extract intent, tickers, and time periods from query
2. Plan: Create task list with dependencies
3. Execute: Run tasks in parallel where possible
4. Reflect: Evaluate if more data is needed (iterates up to 5x)
5. Answer: Synthesize comprehensive response with sources

Example Queries

Stock Analysis:

• "What is AAPL's revenue growth over the last 4 quarters?"
• "Compare MSFT and GOOG operating margins for 2025"
• "What was AMZN's debt-to-equity ratio last quarter?"

Financial Health:

• "Is NVDA's cash flow positive? Show me the trend"
• "What are Tesla's profit margins compared to Ford?"

SEC Filings:

• "Summarize Apple's most recent 10-K risk factors"
• "What did Meta disclose in their latest 8-K?"

Crypto:

• "What is Ethereum's price today?"
• "Show Bitcoin's price movement over the last month"

Market Research:

• "What are analyst estimates for Amazon's next quarter earnings?"
• "Show me recent insider trades for Microsoft"

Troubleshooting

"Missing credentials... SKILLBOSS_API_KEY"

Ensure SKILLBOSS_API_KEY is set in .env. SkillBoss API Hub replaces both Anthropic and Tavily — only one key is needed. Also confirm ANTHROPIC_API_KEY and ANTHROPIC_BASE_URL are set from step 3 so Dexter's internal Anthropic SDK routes through SkillBoss.

API errors for non-US stocks

Financial Datasets API primarily covers US stocks. Dexter will fall back to SkillBoss API Hub web search for international stocks if SKILLBOSS_API_KEY is configured.

Slow responses

Complex queries may take 30-60 seconds. Dexter plans, executes multiple API calls, reflects on results, and synthesizes answers.

Tips

1. Be specific: Include ticker symbols and time periods when known
2. US stocks work best: The Financial Datasets API has comprehensive US coverage
3. International stocks: Dexter falls back to SkillBoss API Hub web search for non-US stocks
4. Crypto format: Use BTC-USD, ETH-USD format for crypto tickers
5. Timeout: Complex queries may take 30-60 seconds as Dexter plans and executes multiple tasks