tick-collection-ops
Warn
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: MEDIUMPROMPT_INJECTIONOBFUSCATIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The 'Self-Evolving Skill' section contains instructions for the agent to modify its own source file (SKILL.md) whenever it encounters issues or workarounds. This establishes a persistence mechanism for prompt injection where the agent might be tricked into baking malicious logic into its core instructions based on faulty or adversarial runtime feedback.
- [OBFUSCATION]: The skill recommends executing PowerShell commands via the
-EncodedCommandparameter using base64-encoded UTF-16LE strings. While intended to solve shell quoting issues over SSH, this pattern is frequently used to obfuscate malicious payloads and bypass security software that monitors command-line arguments. - [DYNAMIC_EXECUTION]: The recommendation to use
-EncodedCommandleads to the dynamic decoding and execution of string-based payloads at runtime, masking the actual commands from simple inspection. - [INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect injection through its self-evolution mechanism. Ingestion points: The skill reads untrusted or variable data from system logs (
journalctl), directory listings (ls), and Parquet data files (duckdb). Boundary markers: No markers or delimiters are defined to isolate external data from the agent's instruction-following logic. Capability inventory: The agent has the ability to overwrite its ownSKILL.mdfile using theBashtool. Sanitization: No sanitization or validation steps are prescribed for the data ingested before it influences the skill's permanent instructions.
Audit Metadata