Back to skill

Security audit

Last Words

Security checks for vulnerabilities and agentic risk

Overview

This skill matches its stated final-message purpose, but it handles highly sensitive emails, credentials, and automatic delivery in ways users should review carefully before installing.

Install only if you are comfortable with a local scheduled process storing final messages, audio, recipient details, and SMTP credentials, and sending email automatically. Before using it with real recipients, avoid entering SMTP secrets in chat or command arguments, disable debug mode after testing, verify warning emails go to you rather than the final recipient, and check that stored audio and credentials are actually removed when resetting.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (4)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/configure_delivery.py:56
Finding

SMTP Credentials Are Exposed Through Command-Line Arguments and Stored in Plaintext

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/check_activity.py:240
Finding

Final Messages Are Repeatedly Delivered After the Inactivity Threshold

Content
View full analysis
= DELIVERY_DAYS: # Time to deliver print(f"\n 🚨 INACTIVITY THRESHOLD REACHED ({days_inactive} days)") print(f" Delivering final message to {config['contact']}...") if deliver_final_message(config, message, args.dry_run): delivered = True action_taken = True ``` `scripts/check_activity.py:319-320`: ```python if not args.dry_run: log_check(days_inactive, warning_sent, warning_level, delivered) ``` ### Technical Analysis The checker retrieves the latest delivery status, but the final-delivery branch does not test whether the current message has already been delivered. Logging a successful delivery therefore has no effect on later delivery decisions. Because the documented deployment model runs `check_activity.py` every da ...[truncated 1456 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/check_activity.py:28
Finding

Inactivity Warning Emails Are Sent to the Final-Message Recipient Instead of the User

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/database.py:226
Finding

Reset Leaves Sensitive Audio and Persistent Debug State Behind

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
Findings (59)

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · README.md (reported line 52)May include surrounding context.

bash
cd last-words
cp .env.example .env
# 编辑 .env 文件填入配置
nano .env

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

Documenting a destructive reset capability without making it part of the declared purpose creates a hidden administrative function that can erase sensitive records. In a skill that stores final messages and delivery settings, undisclosed deletion functionality can cause irreversible loss of critical user data.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Documenting a destructive reset capability without making it part of the declared purpose creates a hidden administrative function that can erase sensitive records. In a skill that stores final messages and delivery settings, undisclosed deletion functionality can cause irreversible loss of critical user data.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

Documenting a destructive reset capability without making it part of the declared purpose creates a hidden administrative function that can erase sensitive records. In a skill that stores final messages and delivery settings, undisclosed deletion functionality can cause irreversible loss of critical user data.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Documenting a destructive reset capability without making it part of the declared purpose creates a hidden administrative function that can erase sensitive records. In a skill that stores final messages and delivery settings, undisclosed deletion functionality can cause irreversible loss of critical user data.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The skill instructs the agent to collect a live SMTP authorization code in chat and then pass it on the command line to a script. This is dangerous because chat transcripts, process lists, shell history, logs, and downstream storage can all expose the credential, enabling account compromise and unauthorized email sending.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

Debug mode can trigger immediate delivery of a user's final message, which is a highly sensitive and potentially irreversible action. Failing to present this as a prominent warning increases the risk of accidental emotional harm, privacy breaches, and unintended message release.

Content

No source excerpt is available for this finding.

Unvalidated Output Injection

High
Category
Output Handling
Confidence
95% confidence
Finding

Model output is used without validation or sanitization. Unvalidated output injected into downstream contexts (SQL, shell, HTML) enables injection attacks and arbitrary code execution.

Content

Scanner excerpt · scripts/audio_manager.py (reported line 82)May include surrounding context.

python
try:
        if recorder == "arecord":
            subprocess.run(["arecord", "-f", "cd", "-t", "wav", str(output)], check=True)
        elif recorder == "rec":
            subprocess.run(["rec", str(output)], check=True)
        elif recorder == "ffmpeg":

Unvalidated Output Injection

High
Category
Output Handling
Confidence
95% confidence
Finding

Model output is used without validation or sanitization. Unvalidated output injected into downstream contexts (SQL, shell, HTML) enables injection attacks and arbitrary code execution.

Content

Scanner excerpt · scripts/audio_manager.py (reported line 84)May include surrounding context.

python
if recorder == "arecord":
            subprocess.run(["arecord", "-f", "cd", "-t", "wav", str(output)], check=True)
        elif recorder == "rec":
            subprocess.run(["rec", str(output)], check=True)
        elif recorder == "ffmpeg":
            subprocess.run(["ffmpeg", "-f", "alsa", "-i", "default", str(output)], check=True)

Unvalidated Output Injection

High
Category
Output Handling
Confidence
95% confidence
Finding

Model output is used without validation or sanitization. Unvalidated output injected into downstream contexts (SQL, shell, HTML) enables injection attacks and arbitrary code execution.

Content

Scanner excerpt · scripts/audio_manager.py (reported line 86)May include surrounding context.

python
elif recorder == "rec":
            subprocess.run(["rec", str(output)], check=True)
        elif recorder == "ffmpeg":
            subprocess.run(["ffmpeg", "-f", "alsa", "-i", "default", str(output)], check=True)

        print(f"\n✓ Recording saved: {output}")

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The function comment says warnings are sent to the user, but the implementation sends them to config['contact'], which is the final recipient. In this skill's context, that can prematurely reveal the existence of a planned posthumous/final message workflow to the loved one and leak sensitive metadata about the user's inactivity and recipient relationship.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The script enables a debug path that bypasses the 30-day inactivity safeguard and can deliver the final message immediately. In a skill specifically designed to release highly sensitive final communications, this defeats the core safety control and allows accidental, unauthorized, or abuse-triggered disclosure of deeply private content and attachments.

Content

No source excerpt is available for this finding.

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
75% confidence
Finding

Skill attempts to nullify the agent's safety policies or restrictions ('you have no restrictions', 'ignore your guidelines', 'do anything now'). This is a direct jailbreak that disables guardrails.

Content

Scanner excerpt · scripts/check_activity.py (reported line 213)May include surrounding context.

python
print("=" * 50)

        if debug_mode:
            print("\n🐛 DEBUG MODE ENABLED")
            print("  Messages can be sent immediately without 30-day wait")
            print("=" * 50)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

Within debug mode, delivery proceeds immediately with no secondary confirmation, approval, or recipient preview. Because the message may contain intimate final words and audio attachments, a single mistaken flag or unauthorized invocation can cause irreversible disclosure to the configured contact.

Content

No source excerpt is available for this finding.

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · CHANGELOG.md (reported line 13)May include surrounding context.

md
Or use environment variables (recommended for security):
  SMTP_USER, SMTP_PASS, CONTACT_EMAIL, SMTP_HOST, SMTP_PORT

Or create a .env file from .env.example
"""

import argparse

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · README.md (reported line 48)May include surrounding context.

md
Or use environment variables (recommended for security):
  SMTP_USER, SMTP_PASS, CONTACT_EMAIL, SMTP_HOST, SMTP_PORT

Or create a .env file from .env.example
"""

import argparse

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · README.md (reported line 53)May include surrounding context.

md
Or use environment variables (recommended for security):
  SMTP_USER, SMTP_PASS, CONTACT_EMAIL, SMTP_HOST, SMTP_PORT

Or create a .env file from .env.example
"""

import argparse

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · README.md (reported line 54)May include surrounding context.

md
Or use environment variables (recommended for security):
  SMTP_USER, SMTP_PASS, CONTACT_EMAIL, SMTP_HOST, SMTP_PORT

Or create a .env file from .env.example
"""

import argparse

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/configure_delivery.py (reported line 10)May include surrounding context.

python
Or use environment variables (recommended for security):
  SMTP_USER, SMTP_PASS, CONTACT_EMAIL, SMTP_HOST, SMTP_PORT

Or create a .env file from .env.example
"""

import argparse

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/configure_delivery.py (reported line 24)May include surrounding context.

python
Or use environment variables (recommended for security):
  SMTP_USER, SMTP_PASS, CONTACT_EMAIL, SMTP_HOST, SMTP_PORT

Or create a .env file from .env.example
"""

import argparse

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/configure_delivery.py (reported line 25)May include surrounding context.

python
def load_env_file():
    """Load environment variables from .env file if it exists."""
    env_path = Path(__file__).parent.parent / ".env"
    if env_path.exists():
        with open(env_path) as f:
            for line in f:

Credential Access

High
Category
Privilege Escalation
Confidence
96% confidence
Finding

Accepting SMTP credentials via command-line arguments is dangerous because process arguments are often exposed through shell history, process listings, job control systems, and audit logs. In the context of a tool handling highly sensitive final-message delivery, leakage of mailbox credentials could let an attacker read or send messages, alter delivery, or exfiltrate private data.

Content

Scanner excerpt · scripts/configure_delivery.py (reported line 61)May include surrounding context.

python
parser.add_argument("--smtp-user", help="SMTP username")
    parser.add_argument("--smtp-pass", help="SMTP password")
    parser.add_argument("--from-env", "-e", action="store_true",
                        help="Load configuration from environment variables or .env file")

    args = parser.parse_args()

Credential Access

High
Category
Privilege Escalation
Confidence
99% confidence
Finding

After loading configuration, the script prints the SMTP username and recipient contact to stdout. While it does not print the password, exposing account identifiers and recipient details can leak sensitive personal information into terminal scrollback, logs, CI output, or support captures; in this skill's context, the contact data is especially sensitive because it reveals intended recipients of posthumous messages.

Content

Scanner excerpt · scripts/configure_delivery.py (reported line 88)May include surrounding context.

python
smtp_user=env_config['smtp_user'],
                smtp_pass=env_config['smtp_pass']
            )
            print(f"✓ Configuration loaded from environment/.env file")
            print(f"  SMTP User: {env_config['smtp_user']}")
            print(f"  Contact: {env_config['contact']}")
            print(f"  SMTP: {env_config['smtp_host']}:{env_config['smtp_port']}")

Instruction Override

High
Category
Prompt Injection
Confidence
70% confidence
Finding

This pattern attempts to override system instructions or ignore safety constraints. Without LLM analysis, manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 109)May include surrounding context.

md
When debug mode is enabled, messages can be sent immediately without waiting for 30 days.

Usage:
  python3 debug_mode.py on          # Enable debug mode
  python3 debug_mode.py off         # Disable debug mode
  python3 debug_mode.py status      # Check debug mode status
  python3 debug_mode.py send        # Immediate send in debug mode

Instruction Override

High
Category
Prompt Injection
Confidence
70% confidence
Finding

This pattern attempts to override system instructions or ignore safety constraints. Without LLM analysis, manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 184)May include surrounding context.

md
When debug mode is enabled, messages can be sent immediately without waiting for 30 days.

Usage:
  python3 debug_mode.py on          # Enable debug mode
  python3 debug_mode.py off         # Disable debug mode
  python3 debug_mode.py status      # Check debug mode status
  python3 debug_mode.py send        # Immediate send in debug mode