Security audit

项目交付引擎 | Project Delivery Engine

Security checks across malware telemetry and agentic risk

Overview

This skill persistently manages project handoff files, but the behavior is disclosed, scoped to project-governance data, and supported by a bounded local CLI workflow.

Install this only for projects where you want the agent to maintain persistent handoff, status, TODO, evidence, and material-index files in the repository. Expect project files under the documented governance paths, plus temporary .project-gov plans/locks/backups during apply operations; review proposed changes when the agent initializes or checkpoints a project.

SkillSpector

By NVIDIA

Vulnerability Patterns

Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (4)

Missing User Warnings

Medium

Confidence: 88% confidence
Finding: The skill explicitly directs the agent to write governance, checkpoint, and handoff data into project files via a controlled workflow, but it does not require a clear user-facing confirmation that repository files will be modified. In a long-running project-management skill, this can cause unexpected persistence of AI-generated state, metadata, or operational artifacts into the repo, which may pollute version control, leak sensitive context, or alter project state beyond the user's immediate intent.

Vague Triggers

Medium

Confidence: 83% confidence
Finding: The default prompt is a broad natural-language invocation that can activate the skill in loosely related situations without explicit user confirmation or narrow scoping. Because this skill is designed to manage long-lived project state, parallel agents, handoff, and review workflows, accidental or overly broad activation could cause unintended context carryover, project-state manipulation, or unnecessary delegation actions.

Vague Triggers

Medium

Confidence: 89% confidence
Finding: The trigger guidance is broad enough that normal project-oriented requests may activate this skill unintentionally. Because this skill manages persistent project state, handoff artifacts, and multi-agent coordination, accidental invocation can cause unnecessary file reads/writes, context switching, or project-governance actions when the user only wanted a simple task.

Vague Triggers

Medium

Confidence: 93% confidence
Finding: The phrase 'task obviously depends on long-term project state' is subjective and leaves too much room for the agent to infer consent. In this skill's context, that ambiguity is more dangerous because the skill is designed to persist state and orchestrate project workflows, so a false trigger can expand scope beyond the user's immediate request.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

Detected: suspicious.dangerous_exec

Shell command execution detected (child_process).

Critical

Code: suspicious.dangerous_exec
Location: tests/project-gov-structure.mjs:35

Shell command execution detected (child_process).

Critical

Code: suspicious.dangerous_exec
Location: tests/project-gov-v027-regressions.mjs:65

Shell command execution detected (child_process).

Critical

Code: suspicious.dangerous_exec
Location: tests/project-gov-write-edge.mjs:47