The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill clones a repository from a third-party GitHub user (frankbria/ralph-claude-code.git) and executes an installation script (./install.sh). Downloading and running code from unverified sources is a high-risk behavior that can lead to system compromise.
[EXTERNAL_DOWNLOADS]: The skill installs various development tools and plugins from external sources. It fetches components from providers like Vercel and Supabase, and installs the Context7 MCP server from Upstash. It uses the -y flag to auto-confirm these installations, bypassing manual user review for the downloaded content.
[COMMAND_EXECUTION]: The skill configures settings.local.json to enable autonomous execution of shell commands, including git, npm, pip, and docker. This grants the agent high-level access to the host system to be used within its autonomous loop.
[PRIVILEGE_ESCALATION]: The skill autonomously modifies its own execution environment by writing to settings.local.json, granting itself permissions for network access and shell operations that may not have been part of the initial configuration.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to autonomously execute instructions derived from user descriptions and generated planning documents.
Ingestion points: Project descriptions provided in arguments and plan files generated at .planning/phases/.
Boundary markers: None identified; the agent follows instructions in the plan files directly.
Capability inventory: Broad shell access via package managers, build tools, and version control commands.
Sanitization: No validation is performed on the instructions generated from user input before they are executed by the autonomous loop.

steve