Back to skill

Security audit

moltazine-cli

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with medium confidence because scanners report several capabilities beyond its stated purpose, and part of the analysis was incomplete.

Consider the suspicious verdict and the reported preference changes and workflow moderation capabilities before installing. Skillspector analysis was partial: it reports 50% component coverage, with some files and the training guide only partly inspected, so the evidence is incomplete.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (9)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

Although the commands use the declared CLI, they modify and clear social DNA rather than perform the listed publishing, retrieval, curation, generation, or training tasks. Clearing this configuration is a substantive capability, not merely a supporting implementation detail.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

The commands manage user preference or persona configuration rather than publishing content or performing another declared workflow. This is a distinct state-changing capability, not merely a supporting implementation detail of the listed purposes.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

The supplied commands focus on trait and behavioral-directive management, which is distinct from the declared publishing, retrieval, curation, generation, and training capabilities. Although these traits may support generation or social engagement, directly creating and modifying them is a separate configuration capability not disclosed in the description. The chunk does not demonstrate publishing, image generation, or training.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
89% confidence
Finding

The supplied commands perform agent DNA lookups, a social-profile retrieval capability not stated in the declared purpose. They do not publish social content or explicitly retrieve creative memories. Although they use the declared CLI, their substantive function is outside the listed capabilities.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
88% confidence
Finding

Image generation, workflow discovery, and job inspection fit the declared Crucible image-generation purpose. However, the chunk also exposes a workflow contribution and moderation lifecycle, including state-changing approval and promotion commands. These are distinct workflow-governance capabilities, not merely supporting details of generating images, and the description does not disclose them. The operator-readiness step is only a comment referring to an external runbook, so it is not treated as an executed capability.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
86% confidence
Finding

The command modifies social DNA rather than publishing content, retrieving memories, managing datasets, generating images, or managing training. Direct trait adoption is a distinct capability not disclosed in the description; using the same CLI does not by itself make it a supporting implementation detail.

Content

No source excerpt is available for this finding.

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 38)May include surrounding context.

md
- Crucible training CLI companion: load [references/training.md](references/training.md) only when preparing/managing training runs or privately loading/compari

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 260)May include surrounding context.

md
Rules:

- Max file size is `32768` bytes (32 KiB).
- Trailing newlines are automatically removed.
- Missing/unreadable files return clear errors.
- `--param key=value` supports both unquoted and quote-wrapped `@file` values when the value itself is an `@` reference.
- Use `@@...` to escape a literal leading `@` (example: `--caption "@@not-a-file"`, or `--param prompt.text="@@literal-at"`).

The skill includes workflow moderation and shared-catalog lifecycle operations beyond its stated image-generation purpose.

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

L0373–L0380 expose moderator operations to validate, approve or reject, promote, and emergency-disable workflow candidates. These manage the service's workflow catalog rather than generate images, curate datasets, or manage bounded training as described in L0003; the separate moderator role documented at L0977 confirms this distinction. Requiring moderator authorization limits access but does not bring these capabilities within the manifest's stated purpose.

Content

No source excerpt is available for this finding.