The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions for the agent to download a shell script from https://sh.jbang.dev and execute it via bash as a fallback installation method. Although the skill mandates that the user review the file content before execution, this pattern constitutes remote code execution from an external source.
[EXTERNAL_DOWNLOADS]: The skill initiates multiple downloads of remote artifacts, including the JBang installer script, the @upstash/context7-mcp package via npx, and the io.quarkus:quarkus-agent-mcp artifact via jbang. These resources originate from third-party domains.
[COMMAND_EXECUTION]: The skill executes various system commands to probe for existing tools (java, docker, mvn, etc.) and uses agent-specific command-line interfaces (claude mcp, bob mcp, codex mcp) to modify agent configurations and register new MCP servers.
[PROMPT_INJECTION]: The skill creates an indirect prompt injection surface by writing persistent convention files (CLAUDE.md or AGENTS.md) into the project root. These files contain mandatory instructions that govern the agent's behavior for all subsequent tasks in the project.
Ingestion points: The skill reads from local templates: templates/conventions-CLAUDE.md and templates/conventions-AGENTS.md.
Boundary markers: The skill uses explicit  and  markers to identify and manage the injected instruction blocks.
Capability inventory: The skill utilizes network access (curl), shell execution (bash), package managers (npx, jbang), and file system modification tools (printf, mkdir).
Sanitization: The skill includes instructions to redact sensitive information and avoid handling API keys in plaintext, though it does not detail specific sanitization for the instruction templates themselves.

setup-agentic-scaffolding