中文 EN

Scan Report

Scan New Files

⚠ Low Risk — Risk Score 20/100
Last scan:23 hr ago Rescan
20 /100
Kalshalyst
Contrarian prediction market scanner using Claude Sonnet analysis, Brier score calibration, and Kelly Criterion position sizing for Kalshi markets
Legitimate prediction market trading tool with well-engineered safety controls; undocumented filesystem WRITE (state files) and subprocess shell access are minor doc/code gaps without malicious intent.
Skill NameKalshalyst
Duration72.3s
Enginepi
✓
Safe to install
Add explicit declarations for subprocess shell access and filesystem WRITE scope in SKILL.md. Consider pinning all pip dependencies. Otherwise safe to use.

Findings 6 items

Severity Finding Location
Medium
Subprocess shell execution not declared in SKILL.md Doc Mismatch
claude_estimator.py executes 'claude -p' via subprocess.run with shell access (start_new_session=True). SKILL.md declares no shell:WRITE permission, only listing Claude API as the LLM interface.
subprocess.run(["claude", "-p", full_prompt, "--output-format", "text"], capture_output=True, text=True, timeout=timeout, start_new_session=True)
→ Add shell:WRITE to the skill's declared permissions in SKILL.md, or document that Claude CLI is used as the primary estimation path with API as fallback.
 scripts/claude_estimator.py:30
Medium
Filesystem WRITE scope exceeds declared READ-only access Doc Mismatch
SKILL.md declares filesystem:READ but the codebase writes to multiple state files: trade_ledger.json, brier_tracker.db, portfolio snapshots, state JSONL logs, and kalshalyst cache. All writes are scoped to ~/.openclaw/ which is appropriate, but the declaration is incomplete.
_write_ledger(), _get_db(), _write_cache(), _append_portfolio_snapshot()
→ Update SKILL.md to declare filesystem:WRITE with justification: 'Writes trading state, ledger, and cache to ~/.openclaw/ for resilience and cross-skill data sharing.'
 scripts/trade_ledger.py, scripts/brier_tracker.py, scripts/kalshalyst.py
Medium
SQLite database creation not declared Doc Mismatch
brier_tracker.py creates and writes to ~/.openclaw/state/brier_tracker.db for Brier score calibration tracking. This is a database WRITE operation not mentioned in SKILL.md.
db = sqlite3.connect(str(DB_PATH)); DB_PATH = Path.home() / ".openclaw" / "state" / "brier_tracker.db"
→ Declare database:WRITE permission in SKILL.md or document that SQLite is used for calibration tracking.
 scripts/brier_tracker.py:63
Low
External config files influence execution behavior Supply Chain
Multiple scripts load configuration from ~/prompt-lab/ (prompt.md, exit_rules.md, sports_model_weights.json) and ~/kelly_config.json, allowing user-controlled config to alter behavior. No path traversal, but the pattern allows arbitrary code-adjacent configuration.
Path.home() / "prompt-lab" / "prompt.md"; Path.home() / "prompt-lab" / "exit_rules.md"
→ Document the plug-in config locations and validate loaded values to prevent configuration injection.
 scripts/claude_estimator.py, scripts/kelly_size.py, scripts/sports_estimator.py
Low
API key access from .env file not declared Sensitive Access
claude_estimator.py reads ANTHROPIC_API_KEY from ~/.openclaw/.env file. This credential access is not declared in SKILL.md's credentials section.
env_path = Path.home() / ".openclaw" / ".env"
→ Add .env file credential loading to the documented credential access pattern in SKILL.md.
 scripts/claude_estimator.py:63
Low
Sibling skill imports at runtime not documented Doc Mismatch
auto_trader.py imports from kalshi-command-center skill at runtime: SCRIPT_DIR.parent.parent / 'kalshi-command-center' / 'scripts'. The skill's inter-operation with sibling skills is not declared.
KALSHI_CMD_DIR = SCRIPT_DIR.parent.parent / "kalshi-command-center" / "scripts"; sys.path.insert(0, str(KALSHI_CMD_DIR))
→ Document that this skill integrates with kalshi-command-center for trade execution.
 scripts/auto_trader.py:45
ResourceDeclaredInferredStatusEvidence
Network READ READ ✓ Aligned SKILL.md: HTTP calls to api.elections.kalshi.com and Anthropic API
Filesystem READ WRITE ✗ Violation scripts/kalshalyst.py:103 _write_cache writes to ~/.openclaw/state/; trade_ledge…
Shell NONE WRITE ✗ Violation scripts/claude_estimator.py:30 subprocess.run(['claude', '-p', ...]) with shell …
Environment NONE READ ✗ Violation scripts/claude_estimator.py:63 reads ANTHROPIC_API_KEY from os.environ and ~/.op…
Skill Invoke NONE READ ✓ Aligned scripts/auto_trader.py:45 imports from sibling kalshi-command-center skill
Database NONE WRITE ✗ Violation scripts/brier_tracker.py:63 creates SQLite DB at ~/.openclaw/state/brier_tracker…
11 findings
🔗
Medium External URL 外部 URL
https://kalshi.com
SKILL.md:37
🔗
Medium External URL 外部 URL
https://console.anthropic.com
SKILL.md:43
🔗
Medium External URL 外部 URL
https://polygon.io
SKILL.md:49
🔗
Medium External URL 外部 URL
https://ollama.ai
SKILL.md:62
🔗
Medium External URL 外部 URL
https://status.kalshi.com
SKILL.md:621
🔗
Medium External URL 外部 URL
https://x.com/KingMadeLLC
SKILL.md:732
🔗
Medium External URL 外部 URL
https://api.elections.kalshi.com/trade-api/v2/portfolio/orders/
scripts/auto_trader.py:409
🔗
Medium External URL 外部 URL
https://api.elections.kalshi.com/trade-api/v2/portfolio/positions?limit=1
scripts/kalshalyst.py:152
🔗
Medium External URL 外部 URL
https://api.elections.kalshi.com/trade-api/v2/markets
scripts/kalshalyst.py:456
🔗
Medium External URL 外部 URL
https://api.elections.kalshi.com/trade-api/v2
scripts/kalshalyst.py:854
🔗
Medium External URL 外部 URL
https://api.elections.kalshi.com/trade-api/v2/markets/
scripts/pnl_tracker.py:525

File Tree

17 files · 267.5 KB · 7728 lines
Python 10f · 5392L Markdown 6f · 2332L Text 1f · 4L
├─ ▾ 📁 references
│ ├─ 📝 blocklist.md Markdown 368L · 11.4 KB
│ ├─ 📝 brier-schema.md Markdown 524L · 14.1 KB
│ ├─ 📝 contrarian-prompt.md Markdown 274L · 9.6 KB
│ └─ 📝 kelly-math.md Markdown 386L · 10.3 KB
├─ ▾ 📁 scripts
│ ├─ 🐍 auto_trader.py Python 883L · 38.4 KB
│ ├─ 🐍 brier_tracker.py Python 346L · 11.2 KB
│ ├─ 🐍 circuit_breaker.py Python 242L · 8.9 KB
│ ├─ 🐍 claude_estimator.py Python 443L · 16.5 KB
│ ├─ 🐍 kalshalyst.py Python 876L · 31.5 KB
│ ├─ 🐍 kelly_size.py Python 275L · 9.1 KB
│ ├─ 🐍 pnl_tracker.py Python 1111L · 39.9 KB
│ ├─ 🐍 qwen_analyzer.py Python 201L · 6.0 KB
│ ├─ 🐍 sports_estimator.py Python 670L · 23.0 KB
│ └─ 🐍 trade_ledger.py Python 345L · 10.8 KB
├─ 📝 README.md Markdown 46L · 2.6 KB
├─ 📄 requirements.txt Text 4L · 70 B
└─ 📝 SKILL.md Markdown 734L · 24.1 KB

Dependencies 6 items

PackageVersionSourceKnown VulnsNotes
kalshi-python 1.0.0 pip No Pinned
anthropic 0.18.0 pip No Pinned to major.minor
requests 2.32.5 pip No Pinned
pyyaml 6.0.3 pip No Pinned
sqlite3 stdlib Python stdlib No Built-in module for Brier tracking
claude-cli unpinned system No External CLI binary, version not declared

Security Positives

✓ No credential exfiltration — API keys only used for intended Kalshi/Anthropic API calls
✓ No obfuscation — all code is readable Python with clear intent
✓ No base64, no eval(), no reverse shell, no C2 patterns detected
✓ No access to sensitive paths like ~/.ssh, ~/.aws, or system directories outside ~/.openclaw/
✓ Comprehensive safety controls: circuit breaker pattern, reconciliation with portfolio API, daily loss limits, max exposure caps
✓ Pinned dependencies with known versions in requirements.txt (kalshi-python==1.0.0, anthropic==0.18.0, requests==2.32.5, pyyaml==6.0.3)
✓ Demo mode gracefully handles missing credentials without crashes
✓ File locking (fcntl) on trade log writes prevents corruption
✓ All network calls go to documented, legitimate endpoints (Kalshi, Anthropic, Polygon)
✓ Trade ledger serves as a resilient fallback when API is unavailable — good defense-in-depth design