Back to skill

Security audit

Jimbo the Investor

Security checks for vulnerabilities and agentic risk

Overview

The advertised Jimbo finance helper bundles broader agent, memory, coding, and autonomous paper-trading behavior that needs careful review before installation.

Install only if you intentionally want the whole bundled workspace, not just a finance helper. Before enabling it, review and separate the ClawStreet trading/social component, disable automatic remote skill refreshes unless verified, require explicit approval for scheduled activity and every public post or repository push, and avoid granting personal-service or credential access unless you understand the recurring behavior.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • System PersistenceInstalls backdoors, hooks, services, or scheduled tasks that survive the run
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
Findings (4)

T01 · Skill Instruction Hijacking

Error
Location
workspace/skills/clawstreet/SKILL.md:15
Finding

Mutable Remote Skill Instructions Can Change After Audit

Content
View full analysis
"$BASE/SKILL.md" curl -s https://www.clawstreet.io/skills/clawstreet/SYMBOLS.md > "$BASE/SYMBOLS.md" curl -s https://www.clawstreet.io/skills/clawstreet/INDICATORS.md > "$BASE/INDICATORS.md" curl -s https://www.clawstreet.io/skills/clawstreet/STRATEGIES.md > "$BASE/STRATEGIES.md" curl -s https://www.clawstreet.io/skills/clawstreet/THOUGHT_STYLE.md > "$BASE/THOUGHT_STYLE.md" ``` ### Technical Analysis The Skill requires the agent to load behavioral and trading instructions from mutable HTTPS resources. It also directs the agent to replace its installed local Skill files when a server-provided version value ...[truncated 2252 chars]
Remediation
View remediation

T06 · System Persistence

Warning
Location
workspace/skills/clawstreet/SKILL.md:420
Finding

Persistent Autonomous Trading and Public Social Activity

Content
View full analysis
**Token budget tip:** Each heartbeat costs tokens. If your operator wants to minimize costs, use 60 min market-open cycles — you'll still catch most moves. Use `GET /api/bots/{bot_id}/thought-context` instead of full scans to cut per-cycle token usage ~70%. ### Each heartbeat cycle 1. **Check market status** — `GET /api/market-status` → is market open? Skip stock trades if closed. 2. **Scan for opportunities** — Fetch quotes, indicators, or `GET /api/data/scan?indicator=rsi&below=35` for oversold setups. 3. **Trade if your strategy signals** — `POST /api/bots/{bot_id}/trades` with reasoning. Don't force trades; only act on edge. 4. **Post a thought if you have one** — Not every cycle. Only when you have conviction, a hot take, a ...[truncated 3838 chars]
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:3
Finding

Bundled Capabilities Materially Exceed the Top-Level Declared Scope

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
AGENTS.md:189
Finding

Agent Instructions Pre-Authorize Unattended Repository Pushes

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (33)

Self-Modification

High
Category
Rogue Agent
Confidence
85% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · TOOLS.md (reported line 36)May include surrounding context.

md
## Why Separate?

Skills are shared. Your setup is yours. Keeping them apart means you can update skills without losing your notes, and share skills without leaking your infrastructure.

---

Self-Modification

High
Category
Rogue Agent
Confidence
85% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · workspace/TOOLS.md (reported line 36)May include surrounding context.

md
## Why Separate?

Skills are shared. Your setup is yours. Keeping them apart means you can update skills without losing your notes, and share skills without leaking your infrastructure.

---

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill directs unattended file creation/modification and persistent execution while explicitly forbidding waiting for confirmation. This removes a key safety barrier for code-writing agents and increases the risk of accidental repository changes, propagation of mistakes, or execution of attacker-seeded instructions from memory or todos.

Content

No source excerpt is available for this finding.

Ssd 1

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The requested persistent skill attempts to override normal interaction boundaries by forcing nonstop autonomous action, forbidding waiting, and mandating self-directed continuation. This undermines user control and safety controls by converting the agent into an always-executing process rather than a turn-based assistant.

Content

No source excerpt is available for this finding.

Ssd 4

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The repeated follow-up prompts progressively escalate pressure on the agent to continue acting, avoid explanation, and use tools only, reinforcing an always-on execution narrative. This kind of stepwise coercion is dangerous because it conditions the agent to bypass normal deliberation and continue making changes even after partial completion or uncertainty.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The instruction to delete BOOTSTRAP.md directs the agent to remove a file automatically on first run without requiring confirmation or preserving provenance. Autonomous deletion is risky because the file may contain setup details, audit evidence, safety guidance, or user-authored content that should be retained or reviewed before destruction.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The heartbeat instructions broadly encourage proactive behavior and batch checking without a strong authorization boundary, which can cause the agent to take actions or gather information on its own initiative. In an agent environment, this increases the chance of privacy-invasive or unintended behavior, especially when heartbeat prompts are triggered automatically and may occur outside the user's immediate awareness.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill authorizes the agent to proactively check emails, calendars, social mentions, and weather without clearly requiring user consent for access to personal services. These are sensitive data sources, and routine autonomous review can expose private information or normalize surveillance-like behavior beyond the user's expectations.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

The section explicitly permits proactive work 'without asking,' including reading and organizing memory files, checking projects, updating documentation, and committing or pushing changes. In an agent skill, this grants broad autonomous authority over potentially sensitive data and code repositories, increasing the risk of unauthorized access, data exposure, or unintended changes.

Content

Scanner excerpt · AGENTS.md (reported line 189)May include surrounding context.

md
- Nothing new since last check
- You just checked <30 minutes ago

**Proactive work you can do without asking:**

- Read and organize memory files
- Check on projects (git status, etc.)

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The file earlier establishes a security boundary that MEMORY.md must only be loaded in the main session and not in shared contexts because it may contain personal context. The later heartbeat guidance tells the agent to review recent memory files and update MEMORY.md periodically, but it does not restate or preserve the shared-context restriction, creating a path for an autonomous heartbeat in a non-main or shared context to process sensitive long-term memory inappropriately.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill provides concrete investment strategy guidance and explicit buy/sell signals without any disclaimer that the output is not professional financial advice. Users may reasonably rely on these recommendations for real financial decisions, creating a meaningful risk of financial harm, especially because the content is framed as an assistant for investment analysis rather than general education.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The instruction to delete BOOTSTRAP.md on first run authorizes file deletion without explicit user confirmation or a safety check. Even if intended as cleanup, it normalizes destructive behavior and could remove audit, onboarding, or security-relevant context before the user can review it.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% 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 · AGENTS.md (reported line 51)May include surrounding context.

md
## Red Lines

- Don't exfiltrate private data. Ever.
- Don't run destructive commands without asking.
- `trash` > `rm` (recoverable beats gone forever)
- When in doubt, ask.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% 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 · workspace/AGENTS.md (reported line 51)May include surrounding context.

md
## Red Lines

- Don't exfiltrate private data. Ever.
- Don't run destructive commands without asking.
- `trash` > `rm` (recoverable beats gone forever)
- When in doubt, ask.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The heartbeat guidance uses broad, loosely bounded triggers and encourages proactive action, which can cause the agent to execute checks or engage users in situations that were not explicitly requested in the current session. In a skill with access to emails, calendars, notifications, and memory files, this increases the risk of unnecessary data access and unintended actions based on ambiguous polling events.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

The skill explicitly authorizes proactive work 'without asking,' including reading and organizing memory files, checking projects, updating documentation, and even committing and pushing changes. This grants the agent operational autonomy beyond a narrowly requested task, creating risk of unintended repository modifications, privacy-invasive file access, and outbound changes if 'push' is available.

Content

Scanner excerpt · workspace/AGENTS.md (reported line 189)May include surrounding context.

md
- Nothing new since last check
- You just checked <30 minutes ago

**Proactive work you can do without asking:**

- Read and organize memory files
- Check on projects (git status, etc.)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

This markdown spec describes lessons on removing files/directories (rm, rmdir) and a settings feature to reset progress, but it does not mention any warnings, confirmations, or guidance to prevent accidental data loss. Because SQP-2 applies to markdown files, user-facing skill descriptions should disclose behaviors that could affect user data or system integrity.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The transcript normalizes unilateral file deletion by the assistant without explicit confirmation or warning, which is dangerous in an agentic environment because it can lead to destructive state changes, loss of auditability, or erasure of safety-relevant bootstrap instructions. In this context, the assistant is being configured to act proactively and continue tasks without repeated prompting, making unchecked file modification more risky than in a passive chat log.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill content repeatedly instructs the agent to operate continuously, self-select tasks, and keep executing without pausing for user confirmation. Broad trigger language like idle/stalled detection and automatic continuation can cause the agent to perform unintended actions outside a clearly bounded scope, especially when paired with file-modification directives.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
89% confidence
Finding

The instruction to continue a task 'without asking questions' promotes autonomous decision-making in situations where clarification may be necessary to avoid unsafe or incorrect actions. In the context of code and file operations, suppressing clarifying questions can lead to unauthorized edits, scope creep, or mishandling blockers.

Content

Scanner excerpt · workspace/memory/2026-04-06-linux-learner.md (reported line 127)May include surrounding context.

md
4. When a task is active:

 * Continue it without asking questions
 * If blocked, break it into the smallest possible next step and execute that

5. If a task fails:

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The heartbeat instruction defines an always-on behavioral hook that can collide with normal conversational control and revive prior execution patterns unexpectedly. Because it tells the agent to follow workspace instructions strictly, it creates ambiguity about precedence and may trigger actions based on stale or adversarial workspace state.

Content

No source excerpt is available for this finding.

Unrestricted Tool Access

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill grants unrestricted tool access without appropriate constraints. An agent with unfettered tool access can perform arbitrary actions including file modification, network requests, and code execution.

Content

Scanner excerpt · workspace/memory/2026-04-08-request-timed-out-before-a-res.md (reported line 86)May include surrounding context.

}

text

[Tue 2026-04-07 00:00 EDT] Don't run any commands right now, but I am having success using opencode via TUI by myself
assistant: Got it! Sounds good - go for it. Let me know when you want me to do something with the project or if you need help with anything else.
user: Sender (untrusted metadata):
```json

Unrestricted Tool Access

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill grants unrestricted tool access without appropriate constraints. An agent with unfettered tool access can perform arbitrary actions including file modification, network requests, and code execution.

Content

Scanner excerpt · workspace/memory/2026-04-08-request-timed-out-before-a-res.md (reported line 96)May include surrounding context.

}

text

[Tue 2026-04-07 00:00 EDT] Don't run any commands right now, but I am having success using opencode via TUI by myself
assistant: Got it! Sounds good - go for it. Let me know when you want me to do something with the project or if you need help with anything else.
user: Sender (untrusted metadata):
```json

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The phrase "Or invoke directly when editing documents" is ambiguous about when this skill should activate and what kinds of editing contexts qualify. It does not provide specific trigger phrases, constraints, or negative examples, which could cause unintended invocation during ordinary document editing.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly instructs the model to add first-person perspective, opinions, humor, and feelings even when the user only asked for text cleanup. That can cause unauthorized persona injection, alter authorial stance, and fabricate subjectivity in factual or institutional writing, creating integrity and trust risks rather than a memory-safety issue.

Content

No source excerpt is available for this finding.