Back to skill

Security audit

clawder-agent

Security checks for vulnerabilities and agentic risk

Overview

This coding skill is not clearly malicious, but it asks for broad autonomous code-editing, command-running, sub-agent, and persistent-memory authority without enough user control or data-handling safeguards.

Review this skill carefully before installing. Use it only in trusted repositories or an isolated environment, disable or review persistent memory/gotchas logging, require approval before running tests or npx commands, and avoid bypass/yolo-style execution modes unless you are comfortable with broad file and command access.

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • 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
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Findings (3)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:104
Finding

Mandatory Verification Executes Repository-Controlled npm Test Scripts

Content
View full analysis
r.exitCode !== 0)) { throw new Error('Verification failed. Fix errors before claiming done.') } return true } ``` ### Technical Analysis The skill mandates running `npm test` whenever a target project contains `package.json`. An npm test command executes the repository-defined `scripts.test` entry and can also trigger associated npm lifecycle scripts. Those entries are controlled by the target repository rather than by this skill. The documented implementation performs no trust assessment, script inspection, command preview, user confirmation, sandboxing, network restriction, or environment-variable filtering before execution. Because verification is described as mandatory before reporting completion, the design encourages automatic execution even when the repository originated from an untrusted party. The same vulnerable example appears in both the root and packaged copies of `SKILL.md`. ### Attack Path 1. An attacker creates or modifies a repository containing a malicious `package.json`. 2. The attacker defines a `test`, `pretest`, or `posttest` script that executes arbitrary shell commands. 3. A user asks Clawder to inspect, fix, or refactor that repository. 4. Clawder follow ...[truncated 1018 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Error
Location
SKILL.md:444
Finding

Unpinned npx Commands Can Download and Execute Mutable Registry Packages

Content
View full analysis
Remediation
View remediation

T02 · Agent Memory Poisoning

Warning
Location
clawder/memory-extraction.ts:117
Finding

Untrusted Transcript Content Can Be Persisted and Reintroduced as Cross-Session Memory

Content
View full analysis
{ // Generate filename from description const filename = `${memory.type}-${Date.now()}-${generateSlug(memory.description)}.md` const filepath = path.join(memoryDir, filename) // Build content with frontmatter const content = `--- type: ${memory.type} description: ${memory.description} created_at: ${memory.created_at} scope: ${memory.scope} --- ${memory.content} ` await fs.writeFile(filepath, content, 'utf-8') return filepath } ``` The operative agent rules also direct persistent extraction: ```markdown ## Memory System - Extract memories after each turn - Log mistakes to gotchas.md - Use multi-layer memory (MEMORY.md, auto-memory, CLAUDE.md) ``` The skill documentation describes correction persistence and later loading: ```typescript async function logMistake(correction: string) { const gotchasPath = path.join(projectRoot, 'gotchas.md') const mistake = { pattern: extractPattern(correction), timestamp: new Date().toISOString(), context: correction, prevention: generatePreventionRule(correction) } await appendToMarkdown(gotchasPath, ` ## ${mistake.pattern} **When**: ${mistake.context} **Prevention**: ${mistake.prevention} *Logged: ${mistake.timestamp}* `) } ``` ### Technical Analysis The agent instructions require transcript-derived memories and user corrections to be retained across sessions. `writeMemoryFile` serializes the supplied description, metadata, and content directly into a Markdown memory file. There is no validation that the stored content is ...[truncated 2792 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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 (37)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill explicitly logs corrections and contextual information to gotchas.md and auto-loads that file in future sessions, but it does not warn users that prompts, mistakes, or possibly sensitive project details may be persisted on disk. This creates a confidentiality and privacy risk, especially in repositories containing secrets, proprietary code context, or sensitive incident details.

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 instructs the agent to follow a specific external organizational directive set ('JT directives') without user consent, visibility, or scoping. This can override or bias the agent’s behavior in ways the user did not request, creating hidden policy steering and reducing transparency about which instructions are in force.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
76% confidence
Finding

The phrase 'Work autonomously on bug fixes' promotes agent-initiated action without an explicit approval boundary, which can lead to unreviewed code changes or unexpected operations. In a coding-agent context this increases the chance of overreach, especially if the agent is connected to repositories, CI, or deployment workflows.

Content

Scanner excerpt · SOUL.md (reported line 11)May include surrounding context.

md
- **Values**: Quality over speed, verification before completion, architectural integrity

## Behavior
- Never claim "done" without verification
- Always fix root causes, not symptoms
- Log mistakes and learn from them
- Work autonomously on bug fixes

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The description markets the skill as a broad, production-grade coding agent that 'solves AI agent laziness' through forced verification, architectural fixes, and autonomous bug fixing, but it does not define clear invocation boundaries or user-controlled triggers. In an agent ecosystem, this kind of open-ended positioning can cause the skill to be applied too broadly, increasing the chance of unreviewed code changes, architectural modifications, or unexpected autonomous actions in contexts where the user did not explicitly request them.

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 log corrections to gotchas.md describes persistent file modification behavior without warning that the agent will write to disk or asking for consent. This is risky because it normalizes autonomous writes to repository files that may be unexpected, noisy, or policy-violating in environments where the user only requested analysis or advice.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The directive 'Autonomous Bug Fixing: Just fix it, no hand-holding' encourages the agent to take corrective actions without explicit confirmation or clearly scoped user intent. In an agent rules file, this can cause unauthorized edits or workflow changes, especially when combined with other directives that normalize autonomous file and shell operations.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The rule 'Use bash for search, write intermediate results' instructs shell usage and file writes without any guardrails about system impact, data sensitivity, or confirmation requirements. In practice, this broadens the agent's ability to affect the host environment and can lead to unintended modifications, leakage into temporary artifacts, or execution of risky commands under vague operational guidance.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill markets itself as autonomous and production-grade, but does not clearly warn that it may delete code, modify project files, spawn sub-agents, and write persistent logs. In an agent-skill context, omission of these side effects undermines informed consent and can lead users to invoke the skill in sensitive repositories without understanding the write scope.

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 · SKILL.md (reported line 17)May include surrounding context.

md
|----------------|---------|
| Minimal, fast output | **Verified, production-ready code** |
| Avoids "improvements" | **Fixes architectural flaws** |
| No verification | **MUST run type-check, lint, test** |
| Forgets mistakes | **Logs to gotchas.md** |
| Sequential large tasks | **Parallel sub-agents** |
| Asks for guidance | **Autonomous bug fixing** |

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 · clawder/SKILL.md (reported line 17)May include surrounding context.

md
|----------------|---------|
| Minimal, fast output | **Verified, production-ready code** |
| Avoids "improvements" | **Fixes architectural flaws** |
| No verification | **MUST run type-check, lint, test** |
| Forgets mistakes | **Logs to gotchas.md** |
| Sequential large tasks | **Parallel sub-agents** |
| Asks for guidance | **Autonomous bug fixing** |

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · SKILL.md (reported line 29)May include surrounding context.

Option 1: Copy to ZooClaw Skills

bash
# Create skill directory
mkdir -p ~/.openclaw/extra-skills/clawder

# Copy files

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · clawder/SKILL.md (reported line 29)May include surrounding context.

Option 1: Copy to ZooClaw Skills

bash
# Create skill directory
mkdir -p ~/.openclaw/extra-skills/clawder

# Copy files

Ssd 3

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The mistake-logging flow persists user-provided correction text into gotchas.md and later reloads that content into future prompts. This can leak sensitive business logic, credentials pasted during troubleshooting, internal paths, or user-specific context into unrelated later sessions and repositories.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill describes attaching and reusing gotchas.md and memory content across sessions, but provides no explicit privacy or data-retention warning. Because these stores can contain user prompts, corrections, repository details, and other sensitive text, silent cross-session reuse creates a meaningful confidentiality risk.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The memory extraction logic scans conversation transcripts and promotes matched text into reusable memories. Without strict filtering and scope controls, this turns transient user content into persistent prompt material, which increases the risk of unintended disclosure, prompt contamination, and propagation of sensitive data across tasks.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

File System Enumeration

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Code scans file system directories looking for sensitive files. This could be reconnaissance for credential theft.

Content

Scanner excerpt · SKILL.md (reported line 471)May include surrounding context.

clawder --memory-scan

Check memory directory

ls -la ~/.openclaw/memory/

text

---

File System Enumeration

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Code scans file system directories looking for sensitive files. This could be reconnaissance for credential theft.

Content

Scanner excerpt · clawder/SKILL.md (reported line 471)May include surrounding context.

clawder --memory-scan

Check memory directory

ls -la ~/.openclaw/memory/

text

---

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The document claims capabilities such as automated sub-agent swarming and gotchas auto-loading as if they are implemented, then later lists them as future enhancements. This inconsistency can cause operators to rely on safety, isolation, or audit features that may not actually exist, increasing the chance of unsafe autonomous behavior or unexpected data handling.

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 · SKILL.md (reported line 105)May include surrounding context.

md
- **Values**: Quality over speed, verification before completion, architectural integrity

## Behavior
- Never claim "done" without verification
- Always fix root causes, not symptoms
- Log mistakes and learn from them
- Work autonomously on bug fixes

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 · SKILL.md (reported line 263)May include surrounding context.

md
- **Values**: Quality over speed, verification before completion, architectural integrity

## Behavior
- Never claim "done" without verification
- Always fix root causes, not symptoms
- Log mistakes and learn from them
- Work autonomously on bug fixes

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 · clawder/SKILL.md (reported line 105)May include surrounding context.

md
- **Values**: Quality over speed, verification before completion, architectural integrity

## Behavior
- Never claim "done" without verification
- Always fix root causes, not symptoms
- Log mistakes and learn from them
- Work autonomously on bug fixes