agentmail Security Audit Report

🔭 Continuously monitored by ClawSecure Watchtower

Source:

SHA-256:

agentmail is an AI agent skill, created by adboio and published at openclaw/skills. ClawSecure audited agentmail across 8 files through the 3-Layer Audit Protocol covering all ten OWASP ASI Top 10 categories, assigning a security score of 5/100 (High Risk). The 13 findings concentrate in Code Injection, Data Exfiltration and Prompt Injection, including Suspicious Pattern: Ignore previous instructions and Suspicious Pattern: while True:. 7 were rated high or critical severity.

Is agentmail safe?

ClawSecure audited agentmail and assigned a security score of 5/100 (High Risk), identifying 13 findings across Code Injection and Data Exfiltration. Review the findings below before installing.

What did ClawSecure find in agentmail?

ClawSecure identified 13 findings in agentmail, concentrated in Code Injection, Data Exfiltration and Prompt Injection. 7 were rated high or critical severity. The most severe include Suspicious Pattern: Ignore previous instructions and Suspicious Pattern: while True:.

How was agentmail audited?

ClawSecure ran agentmail through its 3-Layer Audit Protocol with full OWASP ASI Top 10 coverage, scanning 8 files from openclaw/skills.

What does a score of 5 mean?

ClawSecure assigned agentmail a security score of 5/100, placing it in the High Risk range. This is driven by 13 findings led by Code Injection that should be addressed before use. ClawSecure derives this score with a weighted deduction model (critical -20, high -10, medium -5, low -2 from a base of 100).

Audit Findings for agentmail

ClawSecure detected 13 security findings in agentmail, spanning Code Injection, Data Exfiltration, Prompt Injection and Resource Abuse.

high · Suspicious Pattern: Ignore previous instructions. Prompt Injection finding detected in SKILL.md:84.
high · Suspicious Pattern: while True:. Resource Abuse finding detected in scripts/check_inbox.py:98.
high · Potentially dangerous code pattern detected: base64.*decode. Code Injection finding detected in SKILL.md.
high · Potentially dangerous code pattern detected: system\(. Code Injection finding detected in EXAMPLES.md.
high · Potentially dangerous code pattern detected: base64.*decode. Code Injection finding detected in EXAMPLES.md.
high · Potentially dangerous code pattern detected: base64.*decode. Code Injection finding detected in WEBHOOKS.md.
high · Potentially dangerous code pattern detected: base64.*decode. Code Injection finding detected in send_email.py.
medium · Environment Variable Enumeration. Data Exfiltration finding detected in scripts/check_inbox.py.
medium · Environment Variable Enumeration. Data Exfiltration finding detected in scripts/send_email.py.
medium · Environment Variable Enumeration. Data Exfiltration finding detected in scripts/setup_webhook.py.
medium · Code block in references/EXAMPLES.md at line 363 contains…. Data Exfiltration finding detected in references/EXAMPLES.md:363.
medium · Missing config.json - agent may not be properly configured. Permissions Manifest finding.

Showing the 12 highest-severity of 13 findings. The full interactive list appears below.

3-Layer Audit Protocol

Security Recommendations for agentmail

Eliminate dynamic code execution

agentmail evaluates code at runtime (for example eval or dynamic exec). Remove dynamic evaluation of untrusted input, and where code generation is unavoidable, sandbox it and validate every input.

Audit external network connections

agentmail sends data to external endpoints. Confirm each destination is expected and authorized, and remove any callback that exfiltrates data. ClawSecure monitors for known exfiltration and C2 endpoints.

Investigate prompt injection vectors

Prompt injection is one of the most critical threats to AI agents. Attackers can embed malicious instructions in content the agent processes, causing unintended actions. Review every point where agentmail processes external content and add input validation and output filtering.

Related Security Research

Why Generic Scanners Fail at AI Agent Security→Beyond Static Scans: Why ClawSecure Verifies Agentic Intent→

Related AI Agent Security Audits

openclaw-dashboard-repoScore 0/100 byterover-cliScore 0/100 explain-code-for-developerScore 0/100 explain-code-for-developerScore 0/100 explain-code-for-developerScore 0/100

Scanned on April 24, 2026. agentmail is one of thousands of agents audited by ClawSecure from the community-curated awesome-openclaw-skills list and the openclaw/skills repository.

Start Your Free Scan