The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from system logs, unit files, and configuration files (e.g., config.yml, journalctl). This creates an ingestion surface where malicious content embedded in these sources could attempt to influence the agent's behavior. The skill mitigates this risk by defining specific processing commands and providing redaction filters to sanitize the data.
[DATA_EXFILTRATION]: The auditing logic requires access to sensitive identity material, including Cloudflare Tunnel tokens and per-tunnel credentials JSON files. The skill mandates 'reaction discipline' where the agent must pipe sensitive outputs through specific sed redaction filters to ensure that raw secrets are never reproduced in reports or logs.
[COMMAND_EXECUTION]: To perform its evaluation, the skill executes common Linux system and network utilities including systemctl, ss, stat, find, and grep. These commands are constrained to read-only operations for evidence collection and state verification.
[EXTERNAL_DOWNLOADS]: The skill references configuration and policy data from official Cloudflare domains (cloudflare.com). These references are used to verify firewall allowlists and edge-layer security settings against well-known, trusted vendor sources.

aegis-cloudflared-tunnel-hardening