The Agent Skills Directory

[PERSISTENCE_MECHANISMS]: The skill provides detailed instructions and script templates for establishing persistence via system schedulers such as crontab and Windows Task Scheduler to automate monitoring loops. Evidence: Bash cron recipe (§11.1) and PowerShell Scheduled Task recipe (§11.2) in SKILL.md.
[PRIVILEGE_ESCALATION]: The provided PowerShell automation recipe explicitly bypasses the system's script execution policy, which is a security control intended to prevent the execution of untrusted scripts. Evidence: Use of the '-ExecutionPolicy Bypass' argument in the Scheduled Task action (§11.2, SKILL.md).
[EXTERNAL_DOWNLOADS]: The skill is designed to fetch data from multiple external security intelligence feeds, including those from GitHub and specialized security monitoring sites. Evidence: Polling endpoints for ransomwatch, ransomware.live, HackerNews, and Reddit (§7.1, SKILL.md).
[INDIRECT_PROMPT_INJECTION]: By processing 'adversary chatter' from public Telegram channels, forums, and RSS feeds, the skill introduces a vulnerability to indirect prompt injection where malicious instructions embedded in the data could influence the agent's context. Ingestion points: Telegram web previews, Reddit RSS feeds, and HackerNews search results (SKILL.md §7.1). Boundary markers: The skill includes instructions to mark such data as 'tentative' and use isolated environments (SKILL.md §5, §7.6). Capability inventory: Leverages 'asm-cli' tools and webhook alert delivery (SKILL.md §11.3). Sanitization: Uses title normalization and classification logic to filter data (SKILL.md §7.3, §9.2).
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill facilitates the automated sending of scan results and security alerts to external webhooks, which could be leveraged to exfiltrate findings if the endpoints are not strictly controlled. Evidence: Webhook payload documentation and alert delivery logic for Slack-compatible endpoints (§11.3, SKILL.md).

continuous-exposure-monitoring