Security audit
Security checks for vulnerabilities and agentic risk
This is a disclosed prompt-orchestration skill made of Markdown templates, with expected workflow risks around sub-agent tracking, file-building agents, and persistent notes but no hidden execution or exfiltration behavior.
Before installing, treat this as a prompt-template and workflow skill rather than executable software. Use narrow output paths, fill in explicit constraints, avoid recording secrets or confidential details in tracking logs, and manually confirm before killing agents or running commands that could modify important files or external systems.
The guidance explicitly tells operators to kill and respawn agents when hung or duplicated, but it does not require checking for unsaved work, side effects, or user approval before termination. In an orchestration skill, this can cause loss of in-progress outputs, duplicate actions, or inconsistent state in external systems if an agent was partway through making changes.
This template explicitly directs a sub-agent to create files, modify code, and run scripts, but it provides no safety guardrails such as requiring explicit user confirmation, constraining writable paths, or warning about execution side effects. In an agent-orchestration context, that omission can lead to unintended filesystem changes or execution of unsafe code generated from ambiguous or adversarial task input.
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.
| No technical constraints | Agent might use incompatible tools |
| Missing output path | Files end up in wrong place |
| No Ralph mode for complex tasks | Agent gives up too early |
| No verification step | Agent declares done without testing |
---
The skill recommends maintaining active-agents.md and LEARNINGS.md with task labels, outcomes, and experiment logs, but it does not caution users against recording secrets, internal project details, customer data, or other sensitive operational metadata. In an orchestration skill, these logs are likely to accumulate meaningful context about workflows and projects, so even seemingly harmless tracking records can become a useful disclosure source if committed, shared, or exposed.
No suspicious patterns detected.