The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The primary installation method involves downloading a shell script from a third-party GitHub repository and piping it directly to bash. This provides the external script with full execution authority on the user's system.
Evidence: curl -fsSL "https://raw.githubusercontent.com/Dicklesworthstone/agentic_coding_flywheel_setup/main/install.sh?$(date +%s)" | bash
Evidence: Similar patterns for specific versions (v0.7.0) and commit SHAs (abc1234).
[PRIVILEGE_ESCALATION]: The skill includes a 'vibe mode' that explicitly modifies system configuration to allow the user to execute administrative commands without a password (passwordless sudo).
Evidence: The 'vibe mode' description states it enables "passwordless sudo for your user" and "all dangerous flags enabled."
[PERSISTENCE]: The skill establishes persistence by modifying shell configuration files and installing system-level services that run automatically.
Evidence: Modifies ~/.zshrc and ~/.acfs/tmux.conf.
Evidence: Installs and manages systemd services for ntm and mcp_agent_mail.
Evidence: Includes instructions for managing SSH authorized_keys for remote access.
[EXTERNAL_DOWNLOADS]: The skill fetches numerous tools and runtimes from external sources at runtime.
Evidence: Fetches configuration and tools from https://raw.githubusercontent.com/Dicklesworthstone/agentic_coding_flywheel_setup/.
Evidence: Downloads and executes the Bun installer from https://bun.sh/install.
Evidence: Downloads and executes the Rust installer from https://sh.rustup.rs.
[CREDENTIALS_UNSAFE]: The skill instructs users to store high-value API keys and access tokens in plain text within shell configuration files.
Evidence: Instructions to echo 'export ANTHROPIC_API_KEY="sk-ant-..."' >> ~/.zshrc.
Evidence: References to environment variables for OpenAI, Gemini, Cloudflare, Supabase, Vercel, and Vault tokens.
[INDIRECT_PROMPT_INJECTION]: The skill creates an inter-agent communication infrastructure (mcp_agent_mail) and session tracking logs (slb) that ingest and process data from multiple AI agents, creating an attack surface where instructions from one agent could influence another.
Ingestion points: mcp_agent_mail inbox, slb activity logs, bv session visualization.
Boundary markers: None specified in the documentation for separating agent-generated instructions from system commands.
Capability inventory: The skill has full file system access, network access (curl/wget), and system service management.
Sanitization: No explicit sanitization or validation of the inter-agent messages is mentioned.

agentic-coding-flywheel-setup