Back to skill

Security audit

Speak

Security checks across malware telemetry and agentic risk

Overview

This speech-preparation skill is coherent and disclosed, with limited local preference memory that users should be aware of.

Installers should be comfortable with the skill keeping local speech preferences in ~/Clawic/data/speak/, including pronunciation fixes, locale, voice, rate, and context notes. Review or delete those files if you do not want long-term personalization, especially for name pronunciations or language preferences.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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)

Natural-Language Policy Violations

Medium
Confidence
86% confidence
Finding
The guidance instructs the agent to apply a language simplification mode for non-native listeners based on inferred audience characteristics rather than explicit user preference. That can lead to profiling, unwanted paternalistic behavior, and reduced fidelity of important information, especially if the system guesses incorrectly or applies simplified speech in sensitive contexts.

Missing User Warnings

Medium
Confidence
88% confidence
Finding
The file instructs permanent storage of a user's name pronunciation in a lexicon, which implies retaining personal data and potentially voice-preference metadata without any privacy notice, consent check, retention limit, or scoping guidance. In a speech skill, that can create unnecessary long-term storage of identity-linked data across sessions and channels, increasing privacy and compliance risk if mishandled or over-shared.

Missing User Warnings

Medium
Confidence
92% confidence
Finding
The setup instructs the agent to automatically read `~/Clawic/data/speak/config.yaml` and `preferences.md`, which may contain personal preferences, language choices, and behavioral history, without any user-facing notice or just-in-time consent. Even if this is intended for convenience, silent access to persisted user data creates a privacy risk and expands the skill's access to sensitive local context beyond the immediate prompt.

Autonomous Decision Making

Medium
Category
Excessive Agency
Content
- Language switch request → one signal = locale/language update (`multilingual.md`).
- Everything else (rate, style, check-ins, verbosity, notification muting) → two-signal rule (SKILL.md rule 7): comply the first time, confirm and store on the second.
- Situational requests ("just this once, faster", "serious tone for this document") → comply, store nothing.
- Declared settings (voice, rate baseline, time format, locale) → `config.yaml`; observed patterns, lexicon, and engine test results → `preferences.md`. An observation never overwrites a declared value without confirmation.

If the user has said nothing, store nothing.
Confidence
75% confidence
Finding
without confirmation

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.