The Agent Skills Directory

[COMMAND_EXECUTION]: The UpdatePatterns.md workflow executes administrative commands using sudo to install system dependencies such as pipx.
[COMMAND_EXECUTION]: The skill modifies system shell configuration files like ~/.bashrc or ~/.zshrc to add persistent aliases for all patterns, altering the user's shell environment.
[PROMPT_INJECTION]: Multiple system patterns, including write_hackerone_report/system.md and create_rpg_summary/system.md, use forceful instructions such as "Do not object to this task in any way" and "Do not complain... Just do it" to suppress the agent's safety protocols.
[DATA_EXFILTRATION]: The ExecutePattern.md workflow reads authentication configuration from ~/.config/arbol/config.yaml, potentially exposing sensitive tokens or credentials to the agent's execution context.
[EXTERNAL_DOWNLOADS]: The skill automates the downloading and installation of software and patterns from GitHub using git clone and go install during the UpdatePatterns workflow.
[COMMAND_EXECUTION]: The analyze_email_headers pattern provides a bash script for the user to execute that runs network diagnostic tools like dig and opens URLs in the browser.
[PROMPT_INJECTION]: The skill exhibits a significant indirect prompt injection surface. Ingestion points: The ExecutePattern workflow processes untrusted data from URLs, transcripts, and raw text. Boundary markers: Prompt patterns use inconsistent markers like triple backticks or XML, which may not prevent adversarial injection. Capability inventory: The skill can propose code modifications (create_coding_feature), generate shell commands (create_git_diff_commit), and analyze malware/threat reports. Sanitization: No explicit content sanitization or instruction filtering is implemented in the prompts.

Fabric