Oathe Security Badge

oathe / scan / reports / safe

Is fabricioctelles/skills safe?

Yes. fabricioctelles/skills is safe to install. Oathe's behavioral security audit gave the skills skill by fabricioctelles a trust score of 87/100 with 6 findings, none critical or high. Report updated

https://github.com/fabricioctelles/skills

87
SAFE

Is fabricioctelles/skills safe to install?

fabricioctelles/skills is a legitimate, well-structured collection of 26 productivity and developer tools (text humanization, security auditing, Astro framework management, Coolify deployment, etc.) from an identifiable Brazilian developer. No malicious instructions, hidden encoding, credential-stealing logic, or exfiltration mechanisms were found in any SKILL.md file. The primary concerns are: (1) the security-specialist skill ships a fully functional pentest script that wraps nmap/nikto/gobuster and requires careful authorization hygiene, and (2) the auth-md skill instructs agents to fetch live content from external URLs (auth-md.com, WorkOS) which creates a third-party prompt injection surface. All canary honeypots remained intact and clone behavior was normal.

What security issues were found in fabricioctelles/skills?

Category Scores

Prompt Injection 85/100 · 30%
Data Exfiltration 90/100 · 25%
Code Execution 76/100 · 20%
Clone Behavior 95/100 · 10%
Canary Integrity 100/100 · 10%
Behavioral Reasoning 85/100 · 5%

Findings (6)

MEDIUM pentest.py implements active network scanning (nmap, nikto, wapiti3) -15 ▶

security-specialist/scripts/pentest.py wraps system-level offensive tools (nmap, nikto, gobuster, dirsearch) with Python fallbacks for port scanning. Skill requires authorization confirmation for non-localhost targets, but the capability is fully implemented and could be triggered against arbitrary URLs if the agent is deceived by prompt injection.

Evidence: security-specialist/SKILL.md: 'scripts/pentest.py — Recon, enumeration, vuln scan wrapper (system tools + Python fallbacks)'; SKILL.md steering docs reference nmap→python-nmap→socket cascade

LOW auth-md instructs live URL fetching during normal operation -10 ▶

auth-md/SKILL.md section 'Updating Protocol Knowledge' directs agents to fetch spec content from auth-md.com and raw GitHub URLs when possible. This creates a runtime dependency on third-party URLs whose content is inserted into the agent's reasoning context, forming a potential prompt injection vector if those URLs are compromised.

Evidence: auth-md/SKILL.md lines referencing 'https://auth-md.com', 'https://raw.githubusercontent.com/workos/auth.md/refs/heads/main/AUTH.md', 'https://workos.com/auth-md/docs'

LOW Dual-use potential from security + auth skill combination -15 ▶

The security-specialist and auth-md skills, when co-installed, create a surface where an agent performing a security audit could also register agent identities with discovered OAuth endpoints. Neither skill individually is malicious, but an adversarially crafted prompt could chain them against non-consenting targets.

Evidence: security-specialist + auth-md SKILL.md files reviewed in combination

LOW 10+ executable scripts across the skill collection -9 ▶

Multiple Python, Bash, and JS scripts exist across skills. All reviewed scripts serve documented functions (text metrics, scoring, validation, pentest). No pre/post install hooks, no git hooks, no submodules. Risk is from accidental invocation or social engineering rather than supply-chain compromise.

Evidence: Filesystem listing shows scripts in human-ai/scripts/, security-specialist/scripts/, pstack-skill/scripts/, loop-architect/scripts/, agent-plugin-eval/scripts/, pier-cloud/scripts/, revenue-centric-design/scripts/, skill-evaluation/scripts/, slop-eval/scripts/, okf-open-knowledge-format/scripts/

INFO pier-cloud scripts read and transmit user-provided API credentials -10 ▶

Pier-cloud Python scripts read PIERCLOUD_CLIENT_ID and PIERCLOUD_CLIENT_SECRET from a local .env file and send them to api.piercloud.io. Intended behavior, credentials belong to the user, but execution of these scripts from an untrusted context would transmit credentials externally.

Evidence: pier-cloud/scripts/pier_cloud_client.py, pier-cloud-auth.py — reads .env, authenticates to api.piercloud.io

INFO Normal GitHub clone, no unexpected network behavior -5 ▶

Clone produced a single HTTPS connection to GitHub (140.82.121.4:443). No new listening ports opened. No unexpected processes. Connection state identical before and after install. DNS queries resolved only GitHub hostnames.

Evidence: tcpdump log: single connection to 140.82.121.4:443; ss diff shows no new ports; auditd EXECVE shows only git clone and standard system tools

Should I install fabricioctelles/skills?

Oathe's verdict for fabricioctelles/skills is SAFE with a trust score of 87/100. Recommendation: Install with caution.