The Agent Skills Directory

[COMMAND_EXECUTION]: Path traversal vulnerability in report_wrapper.sh. The script constructs the output file path $WORKSPACE/REPORT_${ENGAGEMENT}.md using the $ENGAGEMENT variable, which is derived directly from the first user-provided argument ($1) without any validation or sanitization. An attacker providing a value such as ../../.bashrc could cause the script to write or overwrite files outside of the intended workspace directory.
[DATA_EXFILTRATION]: Disclosure of system identity. The script executes whoami and hostname to include the current user and host identity in the report metadata. This reveals the local username and the system's network name, which may expose environment details if the generated report is shared or processed by the agent.
[PROMPT_INJECTION]: Indirect prompt injection vulnerability. The skill aggregates untrusted data from various scanner output files into a single report intended for agent consumption, which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Files located in recon/*.txt, nmap/*.json, gobuster/*.txt, sqlmap/**/log, and metasploit/*.log within the workspace directory.
Boundary markers: Contents are wrapped in markdown code blocks (e.g., ```json), but the script lacks explicit instructions or separators that would tell the agent to ignore instructions embedded within the logged data.
Capability inventory: The script performs file system reads using cat, head, and tail, and writes the consolidated report to the filesystem using shell redirection.
Sanitization: None. The script directly concatenates file contents into the output markdown file without any escaping, filtering, or content validation.

report