Meeting Transcriber

• Outcome: Pass
• Latest audit: May 27, 2026 · 10:56 PM UTC
• Version: 1.0.0

Overview

This skill is a local meeting transcriber that mostly matches its stated purpose, but users should understand it records audio, stores transcripts, and has a setup helper that writes local configuration and support files.

Install only if you trust the local Windows/Conda setup and the external D:\dev\python\voiceFunAsr transcription script it runs. Before recording, get consent from meeting participants, and remember that transcripts may be saved locally and printed into the OpenClaw/terminal output. If you run conda_setup.py, expect it to create ~/.openclaw configuration and local helper files, and review any changes it makes to skill files.

SkillSpector

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Behavioral AST: exec() Call, eval() Call, Dynamic Import
• 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 (7)

subprocess module call

Medium

• Category: Dangerous Code Execution

Content

"python_path": python_path,
            "env_path": env_info.get('audioProject_path'),
            "conda_version": env_info.get('conda_version'),
            "detected_at": subprocess.run(
                'date /T && time /T',
                shell=True,
                capture_output=True,

• Confidence: 92% confidence
• Finding: This code invokes a shell with `shell=True` to run `date /T && time /T`, which needlessly expands the attack surface and can execute unexpected programs if shell resolution or environment state is manipulated. Although the command string is static, using a shell in a setup script is dangerous because it normalizes unsafe process-spawning patterns and can be abused in compromised environments.

Intent-Code Divergence

High

• Confidence: 98% confidence
• Finding: The script presents itself as Conda configuration, but it also modifies `SKILL.md`, writes a startup batch file, and injects code into `meeting_minutes.py`. That mismatch between stated purpose and actual behavior is risky because it performs persistent code changes to the skill implementation, increasing the chance of unauthorized functionality changes or hidden backdoor-like persistence.

Missing User Warnings

Medium

• Confidence: 96% confidence
• Finding: The skill performs live microphone capture and stores meeting transcripts to disk, but the description does not present a clear, explicit privacy warning or consent notice near the primary functionality. This creates a real privacy and compliance risk because users or bystanders may be recorded and sensitive meeting content may be retained locally without informed awareness.

Missing User Warnings

Medium

• Confidence: 97% confidence
• Finding: This portion rewrites `SKILL.md` and later modifies the main skill script without obtaining prior, specific consent for those file edits. Silent source rewriting is dangerous in a skill context because it creates persistent changes that may alter execution behavior, obscure provenance, and make later review difficult.

Missing User Warnings

Low

• Confidence: 90% confidence
• Finding: The script writes a configuration file under the user's home directory containing environment paths and project details without clearly disclosing that persistence before creation. While the contents are not highly sensitive secrets, unexpected persistence of local path metadata can still leak workstation structure and violates least surprise.

Ssd 3

Medium

• Confidence: 95% confidence
• Finding: The record-listing function reads transcript files and prints preview content directly to stdout without any sensitivity classification, confirmation, or redaction. Meeting transcripts commonly contain confidential business discussions, credentials, personal data, or regulated information, so even a preview creates an unintended disclosure channel.

Ssd 3

High

• Confidence: 99% confidence
• Finding: The latest-note viewer prints the full meeting transcript to stdout before asking whether to open the file externally. This creates a clear data exposure path through logs, terminal history, or agent output channels, which is especially dangerous because meeting notes are likely to contain sensitive internal information.

Static analysis

No suspicious patterns detected.