The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the setup of an autonomous loop that uses the --dangerously-skip-permissions flag with the Claude CLI. This configuration allows the agent to execute any tool, including arbitrary bash commands and file modifications, without requiring human approval or review for each action.
[COMMAND_EXECUTION]: The script scripts/claudius_runner.py utilizes subprocess.Popen to programmatically launch and control the Claude CLI. While used to implement the loop logic, this pattern allows the runner script to execute shell commands with high-level privileges.
[EXTERNAL_DOWNLOADS]: The documentation and runner script reference https://claude.ai/install.sh for installing the required CLI tool via a piped bash command (curl | bash). This is a reference to a well-known service provided by Anthropic.
[PROMPT_INJECTION]: The skill implements an autonomous workflow that reads instructions and task lists from locally stored files such as PRD.md and progress.txt. This creates a surface for indirect prompt injection where malicious content placed in these files could influence the agent's behavior while it operates with bypassed security prompts.
Ingestion points: Instructions are pulled from PRD.md and progress.txt as defined in scripts/claudius_runner.py and templates/SYSTEM_PROMPT.md.
Boundary markers: No explicit boundary markers or instruction-isolation delimiters are implemented for the ingested file content.
Capability inventory: The autonomous agent is granted broad capabilities, including file editing, shell command execution (via bash), and git operations (commit/push), all without user confirmation.
Sanitization: No validation or sanitization is performed on the content of the task files before they are processed as instructions by the agent.

claudius-maximus