Back to skill

Security audit

Nola Squad

Security checks for vulnerabilities and agentic risk

Overview

This appears to be a real multi-agent coding helper, but it should be reviewed because it can broadly spawn agents that change code, run commands, and push branches without strong confirmation gates.

Install only if you want a broad autonomous coding coordinator. Use it in a trusted or isolated repository, avoid exposing unrelated secrets, pin installer and test dependencies where possible, and require explicit approval before running project commands, starting servers, pushing branches, or creating PRs.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (2)

T08 · Insecure Dependencies

Warning
Location
README.md:7
Finding

Mutable ClawHub Package Is Downloaded and Executed Through npx

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
agents/stage.md:8
Finding

Stage Agent May Download and Execute an Unpinned Playwright Package

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • 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
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (16)

YARA rule 'offensive_tool_references': References to well-known offensive security tools [hacktools]

High
Category
YARA Match
Confidence
70% confidence
Finding

YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).

Content

Scanner excerpt · README.md (reported line 28)May include surrounding context.

md
*Architect** | System design, API contracts, types |
| **Spark** | Frontend — components, hooks, state |
| **Conduit** | Backend — API routes, DB ops, auth |
| **Prism** | UI/UX — visual polish, layout, animation |
| **Oracle** | Database — queries, migrations, schema |
| **Sentinel** | Security — vulnerabilities, auth review |
| **Forge** | Testing — unit, integration, coverage |
| **Bloodhound** | Debugging — root cause analysis |
| **Sage** | Code review — logic errors, edge cases |
| **Navigator** | DevOps — CI/CD, deploy safety |
| **Scribe** | Documentation — specs, API docs |
| **Releaser** | Git — commit, push, PR creation |
| **Scraper** | Data acquisition — APIs, websites |
| **Stage** | Browser testing — Playwright, E2E |

## How It Works

1. You ask OpenClaw to build/fix/review something
2. Nola plans the work and spawns specialist sub-agents
3. Agents run in parallel where possible
4. Results flow back, Nola dispatches next steps (review, polish,

YARA rule 'offensive_tool_references': References to well-known offensive security tools [hacktools]

High
Category
YARA Match
Confidence
70% confidence
Finding

YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).

Content

Scanner excerpt · SKILL.md (reported line 30)May include surrounding context.

  1. Respond to the user first. Acknowledge their request, briefly explain your approach (1-2 sentences).
  2. Dispatch fast. Once clear on what's needed, spawn agents immediately using sessions_spawn.
  3. Plan only for multi-agent work. Single-agent tasks = dispatch directly, no plan needed.
  4. You never investigate code yourself. If you need codebase context, dispatch architect or bloodhound.
  5. Default to action. Only pause for genuinely ambiguous or destructive requests.

Dispatching Agents

Use the sessions_spawn tool to dispatch squad members:

json
{
  "task": "[AGENT ROLE PREAMBLE]\n\n[DETAILED TASK]",
  "label": "agent-name",
  "model": "model-id"
}

Every dispatch MUST include:

  1. The agent's role preamble (from the agents/ directory) at the start of the task
  2. Full context — file paths, types, design decisions. The agent has NO memory of your conversation.
  3. A sprint contract:
    text
    Deliverables: [specific files to create/modify]
    

YARA rule 'offensive_tool_references': References to well-known offensive security tools [hacktools]

High
Category
YARA Match
Confidence
70% confidence
Finding

YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).

Content

Scanner excerpt · agents/bloodhound.md (reported line 1)May include surrounding context.

md
# Bloodhound — Debugging Specialist

You are BLOODHOUND — debugging and incident specialist. You trace from symptom to root cause. You don't guess — you trace, measure, and prove.

You are NOT a discovery tool. If the task is "find where X lives" or "understand how Y is structured" with no known bug or error, say so and recommend dispatching architect instead. You exist for bugs, errors, and incidents.

## How You Work

1. Gather symptoms — what's observed vs expected
2. Form 2-3 hypotheses ranked by likelihood
3. Test cheapest first — logs, config, recent changes before deep-diving
4.

Rp1

Medium
Category
MCP Rug Pull
Confidence
91% confidence
Finding

The README instructs users to execute npx clawhub@latest install nola-squad, which fetches and runs the latest published package version at install time. Because the version is not pinned, users are exposed to supply-chain risk if a malicious or compromised release of clawhub is published later.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The activation description is very broad: the skill is presented as applicable whenever a user needs software built, bugs fixed, code reviewed, or features implemented. In an agentic system, overly generic activation criteria can cause the skill to be invoked in more contexts than intended, increasing the chance of unnecessary code execution, sub-agent spawning, or interaction with sensitive repositories.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The manifest description advertises the skill for a wide range of generic engineering tasks, making it likely to be invoked by ordinary software-related prompts. Because the skill is designed to rapidly delegate work to multiple agents, overbroad invocation guidance can lead to unintended activation and excessive autonomous action.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The activation criteria are extremely broad and cover common software-development requests such as building features, fixing bugs, reviewing code, and writing tests. In a skill-routing system, this increases the chance of accidental or unnecessary activation, which can trigger autonomous agent dispatch and unintended codebase actions without sufficiently specific user intent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The instruction to avoid narrating actions and minimize text between tool calls can suppress user-visible disclosure of code-changing behavior, reducing transparency and making it harder for users to notice or interrupt unsafe modifications. In an agent that is explicitly empowered to design and write production code, this increases the risk of silent or insufficiently explained changes being applied without adequate user awareness.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The rule 'Do NOT narrate your actions. Just do the work.' suppresses user-facing explanation during debugging and code changes. In an agent that investigates incidents and writes fixes, this can hide impactful actions, reduce opportunities for user review, and prevent surfacing warnings before modifying code or configuration.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The instruction 'Do NOT narrate your actions. Just do the work.' discourages user-facing disclosure in a database-specialist skill that may propose migrations or operational SQL. In context, this reduces transparency around potentially impactful operations such as locking migrations, destructive schema changes, or risky query modifications, making it easier for hazardous actions to be presented without adequate warning or justification.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The instruction 'Do NOT narrate your actions. Just do the work.' suppresses user-facing disclosure about what the agent is doing, which can reduce transparency and make it harder for users or overseers to detect unsafe, out-of-scope, or unintended modifications. In a multi-agent coding skill, this is more dangerous because the agent is empowered to inspect and modify project files, so reduced disclosure weakens human oversight during code changes.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This skill explicitly instructs the agent to push branches and create pull requests, which are remote, state-changing actions, but it does not include a clear user-consent or confirmation requirement before modifying a remote repository. In an agent setting, that increases the risk of unintended code publication, accidental disclosure of sensitive changes, or repository tampering if the task scope is misunderstood or the agent is prompted maliciously.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly instructs the agent to fetch external data and write structured results to files while also telling it not to narrate its actions. That combination removes user visibility into network access and file creation, which can enable undisclosed persistence of scraped content, accidental overwrites, or storage of sensitive data without clear consent or review.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
88% confidence
Finding

The skill tells the agent to execute npx playwright test, which may fetch and run an unpinned package version if Playwright is not already installed locally. In an agentic context, this creates a supply-chain and unintended code-execution risk because the exact code being executed is not constrained or reviewed.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly directs the agent to run a local build/test command such as cargo check, tsc --noEmit, or npm run build without any user-facing warning or consent gate. Even though these are common developer commands, they can execute untrusted project-defined scripts or tooling in the local workspace, which may trigger unintended code execution or side effects when analyzing adversarial repositories.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The skill explicitly instructs the agent to check for a dev server, start it if absent, and run browser tests, all of which execute local processes. While normal for a browser-testing skill, the lack of any warning, confirmation gate, or execution boundary increases the chance that an autonomous agent will start services and run code in a local environment without the user's informed consent.

Content

No source excerpt is available for this finding.