The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Instructions to clone a third-party repository from GitHub (https://github.com/aattaran/deepclaude.git) that is not associated with a recognized trusted organization.
[REMOTE_CODE_EXECUTION]: Instructions to grant execution permissions (chmod +x) and run the downloaded script (deepclaude.sh), enabling the execution of code from an external source.
[COMMAND_EXECUTION]: Requests the use of sudo to create a symbolic link in /usr/local/bin, which installs a third-party script with elevated system privileges.
[COMMAND_EXECUTION]: Modifies the user's shell configuration profile (~/.bashrc) to append environment variables and store API keys, which serves as a persistence mechanism on a sensitive configuration file.
[EXTERNAL_DOWNLOADS]: Fetches the official @anthropic-ai/claude-code package from the public registry.
[PROMPT_INJECTION]: The skill implements custom slash commands in ~/.claude/commands/ that create an attack surface for indirect instructions.
Ingestion points: Configuration files stored in ~/.claude/commands/ and read by the agent.
Boundary markers: Absent; no delimiters or safety warnings are included for the shell command interpolation.
Capability inventory: Execution of curl commands to interact with a local proxy, file writes via echo and sudo, and system modifications via chmod.
Sanitization: Absent; raw shell commands are interpolated directly from processed markdown files.

deepclaude-proxy