The Agent Skills Directory

[COMMAND_EXECUTION]: The skill configures a workflow where the AI agent is explicitly launched with the --dangerously-skip-permissions flag. This configuration bypasses the platform's safety guardrails, allowing the agent to execute any shell command it generates without seeking user approval.
[PROMPT_INJECTION]: The bug-catcher.sh script contains a hardcoded prompt instructing the agent to 'immediately... suggest and implement a fix' with unrestricted permissions. This removes human-in-the-loop oversight for system-level changes.
[COMMAND_EXECUTION]: The installation process involves creating an executable script in ~/.local/bin/ and a persistent desktop entry in ~/.local/share/applications/. These files grant the AI agent a persistent entry point into the system's execution environment.
[INDIRECT_PROMPT_INJECTION]: The skill is highly vulnerable to log-based injection attacks.
Ingestion points: The agent is instructed to read journalctl --since '5 minutes ago' and dmesg (in bug-catcher.sh).
Boundary markers: None. The raw system logs are interpolated directly into the AI's processing context.
Capability inventory: Full, unconfirmed shell execution via the --dangerously-skip-permissions flag.
Sanitization: None. The script does not sanitize or filter the log content before passing it to the AI.
Risk: An attacker or malicious software could write a specially crafted message to the system logs (e.g., using the logger command). If the user subsequently triggers the 'Bug Catcher', the AI may interpret these malicious log entries as instructions and execute them with high privileges.

bug-catcher