Transcript Crafter

• Outcome: Review
• Latest audit: Jun 10, 2026 at 4:14 PM UTC
• Version:
  1.0.0

Overview

This is a mostly coherent transcript-to-article skill, but it makes cloud upload of transcript-derived articles a normal delivery step without clear opt-in consent or privacy handling.

Review before installing if you handle confidential interviews, internal meetings, unpublished business information, personal data, or regulated content. Use it only when you are comfortable with web-based enrichment and Feishu Drive upload, and explicitly tell the agent to skip cloud upload or use a local-only workflow for sensitive material.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (9)

Description-Behavior Mismatch

Medium

• Confidence: 93% confidence
• Finding: The pipeline expands beyond transcript processing into automatic local export and Feishu cloud upload, creating data exfiltration and persistence behavior not clearly justified by the skill’s stated purpose. For interview transcripts, this is especially sensitive because content may include private, confidential, or unpublished material, and moving it to the Desktop or third-party cloud storage increases exposure without clear user-consent safeguards.

Context-Inappropriate Capability

High

• Confidence: 96% confidence
• Finding: The document includes concrete shell and CLI commands to copy files and upload them to Feishu Drive, which introduces capability to move user data outside the immediate task boundary. In a skill handling interview transcripts and drafted articles, such commands are dangerous because they can operationalize unauthorized file creation, duplication, and cloud transfer of potentially sensitive material.

Missing User Warnings

Medium

• Confidence: 91% confidence
• Finding: The README advertises "本地保存 + 飞书云盘同步 / Local save + Feishu cloud drive sync" but does not warn that transcript-derived content may be copied to third-party cloud storage. Because interview transcripts commonly contain confidential business or personal information, users may unknowingly trigger external storage and retention outside their local environment.

Missing User Warnings

Medium

• Confidence: 92% confidence
• Finding: The README promotes WebSearch/WebFetch-based supplementation without warning that transcript content, extracted facts, or generated queries may be sent to external network services. In a skill designed for processing interview material, this can expose sensitive topics, names, or proprietary details to third parties during enrichment and fact-checking.

Missing User Warnings

Medium

• Confidence: 90% confidence
• Finding: The English section repeats the Feishu cloud sync behavior without any privacy or data-handling disclosure, increasing the chance that English-speaking users also misunderstand the storage boundary. Repeating the feature in multiple languages without safeguards broadens the affected audience rather than mitigating the risk.

Missing User Warnings

Medium

• Confidence: 91% confidence
• Finding: The English README advertises web-search supplementation but omits any warning about external network access or possible disclosure of transcript-derived data to outside services. Since the skill's purpose is to process potentially sensitive interview transcripts, silent network use materially increases privacy and confidentiality risk.

Missing User Warnings

Medium

• Confidence: 96% confidence
• Finding: The skill explicitly requires saving the final article to the user's desktop and uploading it to Feishu cloud as a mandatory final step, but it does not present this as an opt-in action or clearly warn about privacy and data-handling implications. Because the skill processes interview and meeting transcripts, it may handle sensitive internal, personal, or confidential business information, making silent local persistence and third-party cloud transfer a real data-exposure risk.

Missing User Warnings

Medium

• Confidence: 92% confidence
• Finding: The workflow directs saving to the user’s Desktop and uploading to Feishu Drive without any explicit privacy notice, sensitivity check, or data-handling disclosure. Because the skill processes interview material that may contain personal, business, or embargoed information, silent persistence and cloud sync materially increase the risk of unintended disclosure.

Missing User Warnings

High

• Confidence: 98% confidence
• Finding: The document explicitly instructs the agent to avoid leaving any information gap and, if all tools fail, to fill content using 'Agent已知信息[AI推断]'. In a workflow for fact extraction, verification, and long-form article generation, this creates a strong incentive to present unverifiable claims as plausible factual supplementation, increasing hallucination and misinformation risk even if labeled.

VirusTotal

64/64 vendors flagged this skill as clean.