emotional-persona

Security checks across malware telemetry and agentic risk

Overview

The skill is mostly coherent and local, but its helper scripts can turn crafted user-controlled memory/search text into local Python code execution.

Install only if you want an agent to keep local emotional memory. Before using the helper scripts, fix or avoid the unsafe python3 -c interpolation issue, and make sure users explicitly agree to what emotional information is stored and how it can be deleted.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Risk analysis

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

#
ASI05: Unexpected Code Execution
High
What this means

A malicious or accidental memory/search string could run commands with the same local permissions as the agent.

Why it was flagged

Free-form user or agent-supplied values are inserted directly into Python source passed to python3 -c. A crafted query or stored observation containing quote sequences could break out of the intended string and execute local Python code.

Skill content
QUERY="${1:?Usage: emotion_memory.sh search <query>}" ... query = '$QUERY'.lower() ... entry = json.loads('''$ENTRY''')
Recommendation

Do not run the helper scripts on untrusted text until fixed. Pass user values to Python via argv, stdin, or JSON files rather than string-interpolating them into Python code; also validate numeric/date arguments.

#
ASI06: Memory and Context Poisoning
Medium
What this means

Private emotional history or personal events may be reused in future interactions and could shape how the agent responds.

Why it was flagged

The skill intentionally stores cross-session emotional patterns and personal context. This is aligned with the stated purpose and appears local, but it is sensitive memory that can affect later responses.

Skill content
Long-Term (cross-session) Store using memory-tools or the provided scripts ... Personal context that matters ("user's dog passed away last month")
Recommendation

Use this only with clear user consent, avoid storing clinical diagnoses or unnecessary sensitive details, review the local memory file periodically, and use the forget function for entries the user wants removed.

#
ASI04: Agentic Supply Chain Vulnerabilities
Low
What this means

The skill may fail unexpectedly or use whatever python3 interpreter is available on the user's machine.

Why it was flagged

The declared requirement only lists bash, while the included scripts call python3. This under-declares a runtime dependency rather than showing malicious behavior.

Skill content
requires":{"bins":["bash"]}
Recommendation

Declare python3 as a required binary or remove the Python dependency from the scripts.

#
ASI09: Human-Agent Trust Exploitation
Low
What this means

Users may over-trust the agent or share more sensitive emotional information than intended.

Why it was flagged

The wording strongly anthropomorphizes the agent. This is consistent with the product concept, but users should understand it is simulated response style and memory, not real feelings or clinical care.

Skill content
Emotional Persona makes agents feel. ... This isn't performance — it's architecture.
Recommendation

Present the persona as a simulation and include clear boundaries, especially for wellness or mental-health-adjacent uses.