The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to download and execute scripts directly from the internet using piping to shell commands.
Evidence: curl -LsSf https://astral.sh/uv/install.sh | sh in SKILL.md.
Evidence: powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex" in SKILL.md.
[EXTERNAL_DOWNLOADS]: The skill installs software directly from an unverified third-party GitHub repository.
Evidence: uv tool install --force git+https://github.com/Alishahryar1/free-claude-code.git in SKILL.md.
Evidence: Fetches the uv tool installation script from Astral's public infrastructure.
[PRIVILEGE_ESCALATION]: The installation instructions for Windows require bypassing the standard script execution policy.
Evidence: Use of -ExecutionPolicy ByPass when running the PowerShell installation script in SKILL.md.
[COMMAND_EXECUTION]: The skill requires the execution of multiple CLI tools to manage the proxy server and local models.
Evidence: Commands like fcc-server, fcc-claude, llama-server, and ollama serve are used to control the local environment.
[INDIRECT_PROMPT_INJECTION]: The proxy server acts as a Man-in-the-Middle, intercepting and transforming messages between the agent and LLM providers, which creates a surface for indirect injection.
Ingestion points: The proxy server intercepts Anthropic Messages API calls at http://localhost:8082/v1/messages (referenced in SKILL.md).
Boundary markers: None identified; instructions do not specify markers to distinguish between system, user, and agent content passing through the proxy.
Capability inventory: The proxy performs network operations to route traffic to external providers such as NVIDIA NIM, OpenRouter, and DeepSeek, as well as local endpoints for Ollama and LM Studio.
Sanitization: The skill states the proxy "strips unsupported parameters" and "converts thinking blocks," but there is no mention of security-focused sanitization or validation of the intercepted content.

free-claude-code-proxy