The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses sensitive system configuration files and cryptographic assets for auditing purposes.
Evidence includes instructions to read /etc/ssh/sshd_config and .ssh/authorized_keys as documented in references/audit-checklist.md and references/ssh-bastion.md.
It describes procedures for collecting system evidence, such as memory snapshots and log archives, in references/incident-runbook.md.
[COMMAND_EXECUTION]: The skill employs various administrative commands to manage system security and network configuration.
It utilizes tools such as nftables, ufw, sysctl, and systemctl.
It invokes system-level utilities like apt, certbot, and fail2ban-client across its operational workflows.
[EXTERNAL_DOWNLOADS]: The skill retrieves configuration data and security tools from well-known external sources.
It fetches IP ranges from Cloudflare to populate firewall rulesets.
It downloads certificate management tools and service-to-service identity providers from Smallstep and ACME.sh.
It references the OWASP Core Rule Set hosted on GitHub for Web Application Firewall tuning.
[PROMPT_INJECTION]: The skill demonstrates an attack surface for indirect prompt injection.
Ingestion points: The agent ingests local reference files and guidance from the references/ directory when prompted by SKILL.md.
Boundary markers: There are no explicit delimiters or 'ignore embedded instructions' warnings used when the agent processes these reference documents.
Capability inventory: Across files like SKILL.md and references/incident-runbook.md, the agent is instructed to use high-privilege shell commands (sudo), modify kernel parameters, and perform network operations.
Sanitization: No mechanisms for validation or escaping of the ingested reference content or resulting command outputs are defined.

network-security