The Agent Skills Directory

[OBFUSCATION]: The skill uses Base64 encoding for dynamically executed Python code in its launcher script to pass multi-line scripts through the shell.
Evidence: In scripts/openghost.sh, the cmd_python function encodes the provided script string using base64 and then executes it within the container using python3 -c "import base64; exec(base64.b64decode('...').decode())".
[DYNAMIC_EXECUTION]: The skill relies on dynamic execution to run its assessment modules and sandbox operations.
Evidence: scripts/openghost-assess.py uses subprocess.run to orchestrate shell commands, and scripts/openghost.sh makes extensive use of docker exec to run allowlisted security tools like nmap, sqlmap, and nuclei.
[COMMAND_EXECUTION]: The skill provides a harness for the agent to execute a wide variety of powerful security and networking tools within a Docker container environment.
Evidence: openghost.sh maintains an ALLOWED_TOOLS list including nmap, sqlmap, nuclei, ffuf, hashcat, and websocat.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch and process untrusted data from remote web targets, creating an attack surface for indirect prompt injection.
Ingestion points: The scripts in scripts/pentest/ (e.g., xss_check.py, bola_check.py) ingest HTTP response bodies and headers from target URLs into the agent's context as evidence.
Boundary markers: The SKILL.md file and autonomous-assessment.md reference instruct the agent to treat automated tool outputs as "leads" or "signals" that require manual validation.
Capability inventory: The skill has the capability to execute shell commands, perform network requests, and write files within the engagement directory.
Sanitization: There is no specific sanitization of ingested content before it is presented to the agent as evidence.

openghost-skill