The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill executes bunx @lucid-agents/cli, which downloads the latest version of a remote package from the npm registry at runtime.
[REMOTE_CODE_EXECUTION]: By using bunx to run an unpinned remote package, the skill performs remote code execution. This allows the tool to run arbitrary code fetched from the internet during the scaffolding process.
[COMMAND_EXECUTION]: User-supplied inputs such as {agent-name} are passed directly into shell commands in Step 2 (bunx @lucid-agents/cli {agent-name}) and Step 6 (gh repo create {agent-name}). This creates a command injection vulnerability where a malicious user could provide input containing shell metacharacters (e.g., ;, &, |) to execute unintended commands.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user requirements and interpolates them into the agent's workflow.
Ingestion points: User input for agent name and endpoints in Step 1.
Boundary markers: None identified.
Capability inventory: The skill has broad capabilities including shell execution (Bash), file system modification (Write, Edit), and network-facing operations via gh and railway-deploy.
Sanitization: There is no evidence of sanitization or validation of the {agent-name} variable before it is used in shell execution strings.

paid-agent