AWS Transform CLI version history

Added the atx custom def learnings subcommand, which opens an interactive TUI session to view and manage the learnings automatically generated for a transformation definition based on previous runs.

Added usage telemetry collection during transformation execution. To disable telemetry, set the ATX_DISABLE_TELEMETRY=true environment variable. For more information, see Usage Telemetry.

Fixed an issue on Windows where file operations using ~ (home directory) paths resolved to an unexpected directory. This caused file read/write failures during a session.

Fixed an issue where MCP server and tool names containing special characters could cause connection failures. The CLI now automatically sanitizes names to use only valid characters.

Added native Windows support. You can now run the AWS Transform Custom commands directly on Windows. (The atx ct commands are not yet supported on Windows)

Added steering inputs to atx ct analysis run so agentic-readiness and modernization-readiness analyses reflect your preferences. Use --context for strategic framing, --agent-scope to set agent access for agentic-readiness, and --prefer/--avoid to steer modernization recommendations toward or away from specific technologies.

Transformation definitions in the skill format can now include a scripts/ folder, letting you bundle helper scripts alongside your SKILL.md and references.

Added --wait to atx ct remediation create, which waits until the remediation completes.

Added headless mode for one-shot execution. Run atx -x "prompt" -t to pass a prompt, get a response, and exit with no interactive prompts. Headless mode requires trusting all tools.

Fixed an interactive-session issue that version 3.4.0 introduced. After the agent finishes responding, the CLI might stay on the "Thinking" prompt (shown with an ellipsis in the UI) instead of returning to the input prompt. The CLI now returns you to the input prompt after each response.

Added atx ct schema, a machine-readable manifest of the full atx ct command surface for scripting and automation.

Added token scope validation when you add or update a source, surfacing credential or permission issues immediately instead of failing later during analysis.

Improved auto-remediation pull requests to include finding context and support draft mode for more reviewable PRs.

Improved CLI progress feedback with real-time context about the current activity, estimated agent minutes, and sub-agent activity.

Added configurable checkpoint commit author identity through the ATX_GIT_COMMITTER_NAME and ATX_GIT_COMMITTER_EMAIL environment variables.

Various small bug fixes and improvements.

Added consistent --json and --help support across atx ct commands.

Fixed security agent setup so the agent space is created during setup and removed on teardown.

Various small bug fixes and improvements.

Added continuous modernization commands (atx ct) to the standard CLI distribution for all users.

Improved sub-agent execution to show a "Thinking..." spinner while waiting for server responses, providing clear visual feedback that work is in progress.

Improved spinner to automatically restart with "Working on it..." after 3 seconds of silence following agent output, so the terminal never appears frozen between processing steps.

Improved thinking spinner with a subtle shimmer animation for a more polished visual appearance.

Added ATX_MAX_VALIDATION_LOOPS environment variable to configure maximum validation loops.

Fixed update check no longer shows false "update available" warnings when the installed version is already newer than the registry version.

Fixed delete --transformation-name "" no longer shows a nonsensical confirmation prompt; empty values are now rejected with a clear validation error.

Fixed unrecognized subcommands now show "unknown command" instead of a confusing internal error message.

Fixed version comparison to use numeric comparison instead of string inequality for update checks.

Added pre-flight transformation name validation that suggests similar names from your registry when a name is not found, helping catch typos without needing a separate list command.

Added optional explanations for shell commands, showing what they do and why.

Improved trust settings banner to include more information about tool permissions.

Added learning-disabled banner when using the do-not-learn flag, giving visibility to the learning feature and how to enable it.

Breaking change: Node.js 22 or later is now required. Upgrade from https://nodejs.org/ if you are using an older version.

Added remote MCP server support — connect to HTTP and HTTPS MCP servers with automatic SSE fallback, Bearer token authentication with environment variable expansion in headers, and proxy support (HTTP_PROXY/HTTPS_PROXY).

Added atx resume subcommand as alias for atx --resume.

Added grep tool for multi-file text search with output control — supports regex patterns, file glob filtering, case-insensitive search, configurable depth/result limits, and output truncation.

Improved MCP implementation to use the standard MCP SDK for more consistent behavior and easier support for future MCP features.

Improved shell output handling — output is now gracefully truncated at 30K characters per stream instead of causing a hard error, preserving the beginning and end of output for context.

Fixed an issue where Ctrl+C during an interactive session could prematurely exit the CLI on Linux/macOS instead of aborting the current operation and returning to the prompt.

Fixed trust rejection message to show a dedicated reason for why trust is unavailable, separate from why the command needed authorization.

Fixed dangerous command warning no longer mentions trusting when the command is in alwaysPromptCommands.

Improved tool descriptions with cross-tool guidance, helping the agent choose the best tool for each task and reducing unnecessary shell usage.

Added alwaysPromptCommands in trust-settings.yaml — configure shell command patterns that always require explicit permission, regardless of trust settings. Supports glob wildcards and matches against each sub-command in compound expressions. trustedShellCommands can override these patterns. Not enforced in non-interactive mode to avoid breaking automated pipelines.

Fixed relative paths not being resolved into absolute paths.

Improved tool approval prompts — each option now shows what it does inline, a “Why” message explains why approval is needed, and trust confirmation includes a hint about trust-settings.yaml for permanent trust.

Changed agent output prefix from > to ATX: for clearer visual distinction between user input and agent output.

Fixed display of resume command when CLI retries are exhausted.

Fixed resume from a different working directory. In interactive mode, prompts for corrected path; in non-interactive mode, fails fast with a descriptive error.

Fixed SSO login failure ("Session token not found or invalid") when the credential provider region differs from the profile's sso_region.

Fixed credential caching so long-running sessions no longer fail when credentials are rotated on disk.

Fixed exec command to exit with code 1 on input validation errors, making the CLI reliable in scripts and CI/CD pipelines.

Added display of warning messages from the service.

Added MFA prompt support for AWS credential profiles.

Fixed STS credential calls to respect HTTP(S) proxy settings.

Added --limit <minutes> option to set an Agent Minutes budget for exec and interactive commands. When the limit is reached, the CLI displays usage and exits with resume instructions.

Fixed logging of the initial user message in conversation logs.

Added support for concurrent task execution, enabling faster transformations.

Added Agent Minutes usage display at the end of each conversation, and /usage command in interactive mode to check usage at any time.

Improved error messages when saving transformation definitions with unsupported files.

Added timeout to service calls to prevent the CLI from hanging when the connection is interrupted.

Fixed CLI process hang after successful completion.

Fixed region validation when resuming conversations.

Removed restriction for MCP server paths — users may now specify any path for the command property in MCP configuration.

Improved stability of MCP server connections.

Added support for proxy configuration via environment variables (https_proxy, no_proxy, etc.).

General stability improvements and bug fixes.