The Agent Skills Directory

[PROMPT_INJECTION]: The file references/bash-agent-notes.md contains instructions directing the agent to disregard common security best practices and heuristics. It explicitly tells the agent to 'Stop the "Eval is Always Bad" Cargo Cult' and to 'avoid knee-jerk reactions' to patterns typically flagged as security risks. This is a directive to override internal safety guidelines for code analysis.
[PRIVILEGE_ESCALATION]: The run_privileged function in references/code-examples.md provides a template for executing commands with sudo. This facilitates acquiring root permissions and performing administrative actions.
[COMMAND_EXECUTION]: The skill documents and provides templates for complex command execution, shell integration, and dynamic command construction, which is the primary functionality of Bash scripting but carries inherent risk.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied data for script development and review, which introduces a potential attack surface.
Ingestion points: Processes user requests to write, create, or modify shell scripts as described in SKILL.md.
Boundary markers: No explicit delimiters or warnings to ignore embedded instructions are present in the provided script templates.
Capability inventory: The skill provides templates for file operations, network interaction via CLI tools, and privilege escalation via sudo.
Sanitization: No sanitization or validation of logic within the scripts is performed.

bash-development