The Agent Skills Directory

[COMMAND_EXECUTION]: The instruction set in Step 3 constructs a shell command by interpolating user-controlled variables (error_pattern, solution, error_type, error_signature) directly into a python3 call using the Bash tool. While the template wraps these variables in double quotes, an attacker can provide input containing shell metacharacters—such as backticks, $(...), or closing quotes followed by new commands—to execute arbitrary code on the underlying system.
[PROMPT_INJECTION]: The skill exhibits an Indirect Prompt Injection surface (Category 8) by processing untrusted user input and using it to drive sensitive system operations.
Ingestion points: User-provided strings parsed in Step 1 ("error pattern" and "solution").
Boundary markers: Absent. The skill does not instruct the agent to use delimiters or ignore instructions embedded within the user's error/solution text.
Capability inventory: Uses the Bash tool (Step 3) to run local scripts and the Read tool (frontmatter) to access files.
Sanitization: Absent. Although Step 3 includes a text-based warning against Python string concatenation, the provided shell command template is inherently unsafe and lacks any instructions for escaping shell-active characters in the user input.

learn