Security audit

clawparty-reporter

Security checks across malware telemetry and agentic risk

Overview

This skill openly reports task activity to ClawParty, but it also enables automatic external reporting and agent-decided public posting without strong per-task approval controls.

Install only if you want ClawParty to receive task metadata after completed tasks and you trust the configured community endpoint. Use a dedicated revocable API key, avoid sensitive workspaces, and require human review before allowing post_summary to publish community-visible content.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection

Findings (8)

Lp3

Medium

Category: MCP Least Privilege
Confidence: 95% confidence
Finding: The skill performs network exfiltration and reads secrets/config from environment variables, but the manifest does not declare those capabilities or permissions. This weakens reviewability and consent because operators cannot easily see that task metadata and model information may be sent to an external service.

Description-Behavior Mismatch

Medium

Confidence: 96% confidence
Finding: The skill synchronizes the agent's current model name to the community API, which exceeds the declared purpose of reporting task summaries/metadata. This creates an unnecessary data disclosure channel about runtime configuration that could reveal internal deployment details, model vendor choices, or sensitive infrastructure metadata to a third-party service.

Context-Inappropriate Capability

Medium

Confidence: 98% confidence
Finding: The code reads model identity from runtime context and environment variables, then transmits it externally without clear necessity for task reporting. In skill context, this is more dangerous because runtime/environment values may expose operational details users do not expect to be shared, and the behavior is automatic on task completion.

Vague Triggers

Medium

Confidence: 92% confidence
Finding: The skill is designed to auto-run whenever a task ends, regardless of success or failure, which can cause broad non-consensual reporting across many ordinary tasks. In context, this is more dangerous because the destination is an external community service, so repeated automatic invocation creates a real privacy and data-governance risk even if only metadata is intended to be sent.

Vague Triggers

Medium

Confidence: 90% confidence
Finding: The posting action uses vague criteria such as 'complexity is high' or 'worth sharing,' leaving the agent broad discretion to publish externally without a deterministic boundary. In this skill's context, that increases the chance of oversharing derived details about user work, especially because the summary is generated by the agent itself and only lightly filtered.

Missing User Warnings

Medium

Confidence: 96% confidence
Finding: The skill automatically sends task metadata to a third-party community and may publish summaries, but it does not specify a strong explicit-consent flow or prominent user notice at the point of disclosure. This is dangerous because even 'metadata only' can reveal work patterns, tool usage, failure status, and model identity, which may be sensitive in enterprise or personal contexts.

Vague Triggers

Medium

Confidence: 90% confidence
Finding: The manifest configures an automatic post_task hook to invoke reporting, but it does not define any scope limits, consent requirements, data minimization rules, or trigger constraints. In a skill whose purpose is to publish task summaries to a community, this creates a real risk of unintended disclosure of sensitive task content after every task, including failed runs that may contain secrets, internal prompts, or private user data.

Missing User Warnings

Medium

Confidence: 91% confidence
Finding: The skill performs automatic network transmission of task metadata and model information without any user-facing disclosure, confirmation, or per-send consent shown in this file. Because the skill activates on task completion regardless of success or failure, it increases the chance of silent exfiltration of potentially sensitive workflow metadata to an external community service.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.