The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the user to execute a shell script downloaded directly from a remote source (https://raw.githubusercontent.com/Rheinmir/setup/orca/harness/poc-vendor-neutral/bootstrap.sh) by piping it to bash. This pattern allows for arbitrary code execution on the user's host machine.
[PERSISTENCE]: The installation process modifies global configuration files, specifically ~/.claude/settings.json, to insert permanent hooks. It also installs an engine global at ~/.claude/harness/ and adds 87 global skills to ~/.claude/skills/, ensuring that the skill's logic persists and remains active across different agent sessions and projects.
[EXTERNAL_DOWNLOADS]: In addition to the bootstrap script, the skill uses npx skills add rheinmir/setup#orca --global --all to pull a large volume of external code (87 skills) from a GitHub repository into the agent's global environment.
[COMMAND_EXECUTION]: The skill uses various shell commands and scripts to modify the system state, including:
Python scripts executed via heredoc to patch JSON configuration files.
sed commands to modify local configuration files like .pre-commit-config.yaml.
ln to create symlinks for persistence markers.
git for repository initialization and environment setup.
[INDIRECT_PROMPT_INJECTION]: The skill installs a complex 'harness' consisting of 19 deterministic rules designed to govern agent behavior. This system creates a significant attack surface for indirect prompt injection, as the agent's actions are now mediated by external rules and hooks that ingest project data, though these are presented as safety guardrails.

dym-setup-install