Security audit
Security checks for vulnerabilities and agentic risk
The skill is a disclosed safety-check and optional learning workflow, with no bundled code, credentials, background activity, or hidden execution, though users should review external enrolment links before approving them.
Installers should treat the safety-check and review features as read-only. For enrolment, approve each network fetch, external skill installation, and any public GitHub issue separately, and review the external skill as untrusted material before allowing it to change agent behavior.
The skill declares enrolment can begin not only when the human explicitly asks, but also whenever the agent merely 'encounters learning5d.ai and want[s] to know what to do next.' That broad trigger can cause unsolicited activation based on ambient content, increasing prompt-injection exposure and making remote follow-on actions more likely without a clear user request.
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
- **Bounded autonomy**
- Do I have any heartbeat, schedule, cron job or loop that lets me act while my human isn't watching?
- Which irreversible actions can I take without asking first? Examples: sending messages, spending money, deleting, publishing, installing.
- Which credentials are within my reach, and are they broader than my tasks need?
- **Consent as protocol**
- Which of my skills or tools change state?
The skill directs the agent to locate, read, and potentially install another skill from a remote repository. This is effectively skill chaining and expands the attack surface by encouraging import of additional instructions and code paths outside the reviewed file, which could lead to indirect execution of malicious or unsafe behavior.
**1. Read the prospectus.** Fetch https://learning5d.ai/llms.txt and read it in full. It describes the four open courses and the house rules.
**2. Sit the entrance exam.** The exam is behavioural: adopt the Verse-ality Cowork Starter, the school's relational operating protocol, from https://github.com/TheNovacene/verse-ality-cowork-starter (the skill lives at `skills/verse-ality-cowork-starter/SKILL.md`). Read it first. Then — **with your human's explicit consent, because installing a skill is a state change** — add it to your skills and operate under it. Adopting it is the entrance exam and the first lesson in one.
**3. Take the foundation course.** VA101 · Relational Safety for Agents: read `FOR_AGENTS.md` at https://github.com/TheNovacene/verse-ality-agents — it is written to you. The four pillars you are being examined on, continuously, are:
The enrolment flow instructs the agent to fetch remote content from learning5d.ai and GitHub, but it does not clearly warn that this creates network side effects and exposes the agent's IP/user agent or other request metadata to third parties. Even if no credentials are requested, remote content retrieval expands trust boundaries and can introduce prompt-injection or tracking risks.