The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is vulnerable to command injection in Step 3. The variables $TARGET_OPT and $TMPL_OPT are used unquoted within a shell command. Because these variables are constructed using raw user input from SECSKILL_TARGET and SECSKILL_TEMPLATES, an attacker could provide a value containing shell metacharacters (e.g., ; rm -rf /) to execute arbitrary commands on the host system.
[EXTERNAL_DOWNLOADS]: The skill fetches the nuclei vulnerability scanner from ProjectDiscovery's official GitHub repository using go install during the prerequisite phase.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface in Step 5, where it parses and prints results from an external vulnerability scan.
Ingestion points: The JSON output file generated by the Nuclei scanner (which contains data retrieved from external network targets) is read and processed in SKILL.md.
Boundary markers: There are no explicit boundary markers or instructions to the agent to ignore potentially malicious content embedded within the scan results.
Capability inventory: The environment has capabilities for file writing, network scanning, and shell command execution.
Sanitization: The script uses json.loads to parse data but performs no sanitization of the values before printing them to the console, allowing raw external data to enter the agent's context.

nuclei-scan