Back in July I wrote about Gaunt Sloth Assistant hitting 0.9.2. Today we finally get to say version 1.0.0 is out. This is the release where we upgraded our primary dependency the LangChain/LangGraph to v1, moved the runtime baseline to Node 24/npm 11, and declared the tool ready for daily automation work.

What changed since the last post?

• Reviews concluded with a call to the built-in rating tool. By default the scale is 10/10, the pass threshold is 6/10, and rates below 6 cause the review command to return non-zero exit codes. If you prefer warnings-only mode, set commands.review.rating.enabled (and/or commands.pr.rating.enabled) to false in .gsloth.config.*.

• Identity profiles are now part of the core workflow, letting you swap prompts, models, and providers per folder with a simple -i profile-name flag.

• Middleware is now first-class. You can stack built-ins such as anthropic-prompt-caching or summarization, or point at your own JS middleware objects, and the CLI shows what runs alongside every command.

• Deep merging for command configs fixes the annoying situation when overriding the content provider deleted the rating settings. Defaults now survive partial overrides.

• OAuth caching, documentation, and the README were refreshed so newcomers can get productive faster, and dependencies were hardened while we were here.

Identity profiles are the everyday quality-of-life feature in 1.0.0. They let me flip between system prompts, model presets, and tool chains per task. gth pr 555 PP-4242 still reads .gsloth/.gsloth-settings, but gth -i devops pr 555 PP-4242 automatically switches to .gsloth/.gsloth-settings/devops/ with whatever prompts and providers that folder declares.

Need to talk to Jira through MCP? Drop a profile such as jira-mcp that contains its own config and call gth -i jira-mcp chat. A trimmed example looks like this:

{
  "llm": {
    "type": "vertexai",
    "model": "gemini-2.5-pro"
  },
  "mcpServers": {
    "jira": {
      "url": "https://mcp.atlassian.com/v1/sse",
      "authProvider": "OAuth",
      "transport": "sse"
    }
  },
  "requirementsProviderConfig": {
    "jira": {
      "cloudId": "YOUR-JIRA-CLOUD-ID-UUID",
      "displayUrl": "https://YOUR-BUSINESS.atlassian.net/browse/"
    }
  },
  "commands": {
    "pr": {
      "contentProvider": "github",
      "requirementsProvider": "jira"
    }
  }
}

Switching between those folders is now just a flag, so I can keep separate personas for DevOps, documentation, or any remote MCP I need to reach.

The rater tool is the other big unlock. Reviews always included qualitative feedback, but 1.0.0 makes the score actionable: we share it with the review module through an artifact store and wire it to setExitCode, so CI can fail automatically when quality is below the goal. Setting guardrails for production services now takes seconds and no longer depends on custom scripts.

Finally, the middleware registry and artifact store give me nicer hooks for future automation. I can wrap model/tool calls, log exactly what ran, and still let Gaunt Sloth handle the chat, code, PR, or init commands it already mastered. The CLI remains a small TypeScript binary you can ship through npm or run via npx gth, but it now has the internal architecture to evolve without hacks.

If you want to try the release, the quickest path is still
npm install -g gaunt-sloth-assistant

The GitHub repo at https://github.com/Galvanized-Pukeko/gaunt-sloth-assistant is there for reference and issues. File an issue, drop feedback in Discussions, or wire the new rater tool into your CI and tell me how it behaves—I would love help pushing 1.1 features.

Huge thanks to all contributors for their PRs and testing.