ssti-server-side-template-injection

Warn

Audited by Gen Agent Trust Hub on Nov 9, 2026

Risk Level: MEDIUMCREDENTIALS_UNSAFEDATA_EXFILTRATIONREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONOBFUSCATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The skill provides instructions and specific payloads to extract highly sensitive credentials, including AWS keys (~/.aws/credentials) and environment variables from /proc/self/environ.\n- [DATA_EXFILTRATION]: Contains multiple payloads for reading system files like /etc/passwd and performing Out-of-Band (OOB) data exfiltration via DNS lookups (nslookup) to attacker-controlled domains.\n- [REMOTE_CODE_EXECUTION]: Provides a comprehensive collection of RCE chains for major template engines such as Jinja2, FreeMarker, Twig, Velocity, and Thymeleaf, enabling arbitrary command execution on the host.\n- [COMMAND_EXECUTION]: Lists numerous dangerous system functions for process execution, including os.popen, subprocess.Popen, system, and exec, used to run shell commands on the underlying system.\n- [OBFUSCATION]: Demonstrates sandbox bypass techniques using hex-encoded strings (e.g., \x5f\x5fglobals\x5f\x5f) to evade static analysis and security filters.\n- [DYNAMIC_EXECUTION]: Details methods for runtime code generation and library injection, such as MRO subclass traversal in Python and Java reflection, to escape template sandboxes.\n- [INDIRECT_PROMPT_INJECTION]:\n
  • Ingestion points: Untrusted data enters the agent context through URL paths, query parameters, HTML forms, and error pages that are subsequently processed by template engines (e.g., render_template_string).\n
  • Boundary markers: Absent; the skill does not use delimiters or instructions to ignore embedded commands in the untrusted data.\n
  • Capability inventory: Extensive capabilities are listed, including shell command execution (system/popen), file system read access, and network operations.\n
  • Sanitization: No input sanitization or validation mechanisms are documented to prevent malicious content in the processed data.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 9, 2026, 12:00 PM