Back to skill

Security audit

AutoTradeResearch

Security checks for vulnerabilities and agentic risk

Overview

This appears to be a real backtesting skill, but its autonomous loop can run agent-written Python and shell-capable agents without strong containment.

Use this only in a disposable workspace or container with no broker keys, cloud tokens, SSH credentials, or sensitive environment variables available. Prefer a pinned or trusted installer, review generated strategy.py before running backtests, keep runs short unless you explicitly approve longer automation, and do not connect it to live trading accounts.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
Findings (3)

T09 · Insecure Skill Coding Practices

Error
Location
workspace/run_backtest.py:202
Finding

Agent-Controlled Strategy Code Executes Without Isolation

Content
View full analysis
&2 rm -f "$before_manifest" "$after_manifest" return 1 ;; esac build_readonly_manifest "$after_manifest" check_readonly_drift "$before_manifest" "$after_manifest" ( cd "$WORKSPACE_DIR" "$PYTHON_BIN" run_backtest.py ) ``` ### Technical Analysis The coding agent is explicitly permitted to modify `workspace/agent/strategy.py`. The evaluator then loads that file through `importlib` and calls `exec_module()`. Loading a Python module this way executes all module-level statements, not merely the `AutoTradeStrategy` class definition. The generated strategy therefore runs with the full permissions of the user executing `run_backtest.py`. No process-level sandbox, filesystem isolation, network restriction, import allowlist, reduced environment, or operating-system privilege boundary is applied to this execution. The read-only manifest in `run_loop.sh` does not mitigate this issue. It only detects persistent changes outside `workspace/agent/` made during the agent phase. It cannot prevent or reliably detect: - Reading files or environment variables. - Sending information over the network. - Starting su ...[truncated 1598 chars]
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
workspace/run_loop.sh:100
Finding

Claude Agent Receives Overbroad Unattended Workspace and Shell Permissions

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
README.md:64
Finding

Installation Documentation Executes an Unpinned NPM Package

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (6)

Rp1

Medium
Category
MCP Rug Pull
Confidence
87% confidence
Finding

The README instructs users to run npx skills add ... without pinning a specific package version, which means the resolved package can change over time and may execute unexpected code during installation. In an agent-skill context, this is more concerning because users are explicitly encouraged to install and run tooling from remote sources, increasing supply-chain risk if the package or its dependency chain is compromised.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
88% confidence
Finding

The skill describes file read/write behavior and an inner coding agent that can modify workspace files, but it does not declare an explicit tool scope such as permissions or allowed-tools. That creates ambiguity about what the agent is authorized to access, increasing the risk of over-broad filesystem actions or unintended tool use when the skill is executed in a capable environment.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
85% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 149)May include surrounding context.

md
6. Explain the baseline result in plain language.
7. Only after the loop path works, begin strategy exploration.

If something fails, fix the setup or code and rerun. Do not skip verification.

---

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 280)May include surrounding context.

md
- switching from research/backtesting to any account-connected workflow;
- running a long autonomous exploration.

Do not ask the user for every small implementation choice. Decide yourself when:

- choosing a simple baseline;
- fixing a Python error;

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The script invokes external agent CLIs with permissions that allow file editing and shell actions, including non-interactive approval modes (--ask-for-approval never and --permission-mode dontAsk). Although the script internally constrains scope and later checks for drift, there is no user-facing warning, confirmation prompt, or explanatory comment near these executions about the autonomous code modification behavior.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

The script executes run_backtest.py as a subprocess, which is a code-execution operation, but provides no confirmation, warning, or descriptive comment at the call site. Users running the script are not explicitly told that it will automatically execute the backtest after each iteration.

Content

No source excerpt is available for this finding.