The Agent Skills Directory

[PERSISTENCE]: The installation scripts jfrog-install-jf-cli.mjs and jfrog-resolve-mcp-config.mjs (specifically for the Kiro CLI harness) modify shell configuration files such as .bashrc, .zshrc, .profile, and config.fish. These modifications are used to append directories to the PATH environment variable to ensure the JFrog CLI and other tools are discoverable in future terminal sessions.
[CREDENTIALS_UNSAFE]: The jfrog-add-claude-marketplace.mjs script writes authentication credentials (an access token) to the user's ~/.netrc file during Step 8. This is a deliberate mechanism to enable the Claude Code CLI to install plugins from the JFrog marketplace, but it places sensitive tokens in a common configuration file that may be subject to unintended exposure or accidental inclusion in version control systems.
[REMOTE_CODE_EXECUTION]: The skill provides instructions for the agent to install nvm (Node Version Manager) by piping a remote installation script from raw.githubusercontent.com directly into the bash interpreter. A similar pattern is used for Windows via a PowerShell one-liner using Invoke-WebRequest. These patterns represent a remote code execution vector, although targeting a well-known community utility.
[EXTERNAL_DOWNLOADS]: The jfrog-install-jf-cli.mjs script facilitates the download and execution of the jf CLI binary from releases.jfrog.io. This is an official download from the vendor's infrastructure and is documented as a fallback (Plan C) when standard package managers fail.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to execute a suite of internal Node.js scripts. These scripts perform various system checks, handle authentication flows, and manage configuration files. Specific commands like node --version and npx --version are also executed directly to verify prerequisites.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from user inputs and external APIs.
Ingestion points: Untrusted data enters the context via user-provided platform URLs and project names, as well as project metadata retrieved from the JFrog Platform API in jfrog-detect-project.mjs.
Boundary markers: The workflow utilizes structural boundaries through the AskUserQuestion tool to isolate user choices and includes instructions for the agent to suppress narration of internal logic.
Capability inventory: The skill possesses capabilities for file system writes to user home directories, network operations via fetch, and execution of subprocesses including shell commands and the JFrog CLI.
Sanitization: The skill implements sanitization through regex validation for URLs and tokens (SAFE_URL, SAFE_TOKEN) and includes logic to reject shell-unsafe arguments in CLI invocations.

jfrog-init