The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by processing untrusted data from external systems to drive autonomous code modifications. Ingestion points: Phase 2 collects error messages, stack traces, and DOM snapshots from external environments that may be attacker-controlled. Boundary markers: The instructions do not specify delimiters or instructions to isolate the collected evidence from the agent's core logic. Capability inventory: The skill utilizes Bash, Write, and Edit tools to modify source code and execute shell commands. Sanitization: There is no evidence of input validation or sanitization of the collected evidence before it is used to plan and execute repairs.
[REMOTE_CODE_EXECUTION]: The skill performs dynamic code generation and execution as its primary function. It is designed to build missing capabilities and patch logic based on external failure context. These generated modifications are subsequently executed through the Bash tool during the verification phase. This self-modifying code pattern creates a path for remote code execution if the generation logic is manipulated by malicious external input.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to autonomously execute verification tests and system operations. The specific commands are determined dynamically based on the failure context and the agent's generated repairs, which increases the risk of command injection if the underlying context is compromised.

self-evolution