The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes utility scripts (goalflow.sh and preflight_audit.py) that execute shell commands such as git and python to perform environment diagnostics and repository security audits. These invocations are safe as they use list-based argument passing in subprocess.run, avoiding shell injection vulnerabilities, and are restricted to auditing the local repository context.
[DYNAMIC_EXECUTION]: The skill provides documentation and auditing for the CodeNode component, which executes Python code via exec(). While dynamic execution is a high-risk capability, the skill explicitly warns that the AST safety guard is currently disabled in the managed framework and instructs users to treat generated inputs as trusted-only, mitigating the risk through transparency and user guidance.
[DATA_EXFILTRATION]: The skill contains a dedicated security auditor (preflight_audit.py) specifically designed to prevent data exposure. It scans for hardcoded credentials, sensitive files in git history (such as .env and .pem files), internal infrastructure IP addresses, and insecure CORS configurations. It provides actionable remediation for secret rotation and history scrubbing.
[PROMPT_INJECTION]: The skill manages a runtime skills engine where Markdown instructions are matched by an LLM and injected into system prompts. It provides a check_goalflow_skill.py tool to audit these injections for size and description quality, helping to prevent both performance degradation and indirect prompt injection by ensuring descriptions are specific and injection costs are visible to the developer.

goalflow