The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill dynamically generates a Bash script (engagement/evidence/spray-runner.sh) by interpolating variables such as target IPs, domains, and service lists provided by the orchestrator. This script is then executed in a single shot using the Bash tool.
[PRIVILEGE_ESCALATION]: Troubleshooting instructions advise the use of sudo ntpdate or sudo rdate to resolve Kerberos clock skew issues, which requires the agent to attempt commands with elevated privileges.
[COMMAND_EXECUTION]: The skill explicitly utilizes dangerouslyDisableSandbox=true when executing its generated spray script via the Bash tool to allow for extended execution times (up to 10 minutes) and background processing.
[INDIRECT_PROMPT_INJECTION]: The skill represents an indirect prompt injection surface by ingesting untrusted data from the orchestrator (usernames, target services, and IPs) and directly interpolating them into a generated shell script without explicit sanitization or validation logic.
Ingestion points: Usernames and targets are passed via the agent prompt from the orchestrator.
Boundary markers: None identified in the script generation template.
Capability inventory: File-write (Write tool) and shell execution (Bash tool).
Sanitization: No visible escaping or validation for shell metacharacters in the interpolated variables.
[CREDENTIALS_UNSAFE]: The core purpose of the skill is the discovery and handling of credentials. It instructs the agent to save valid credentials to local files (e.g., engagement/evidence/spray-results.txt) and manage them via an MCP state server.

password-spraying