@rudderhq/cli 0.6.4 → 0.6.5-canary.1

package/README.md

@@ -1,10 +1,12 @@
+![Rudder — Build your self-improving Agent Team.](docs/images/rudder-social-card.png)
+
 # Rudder
 
 > Build your self-improving Agent Team.
 
 Agents that think, build, play, and learn from real work.
 
-[Website](https://rudderhq.dev) | [Docs](https://docs.rudderhq.dev)
+[Website](https://rudderhq.dev) | [Docs](https://docs.rudderhq.dev) | [Discord](https://discord.gg/ZcfWwPVkUz)
 
 Rudder turns goals, tasks, chats, issues, agent runs, reviews, and feedback into a work loop for agent teams. It gives humans and agents a shared operating structure for moving work forward, running agents, reviewing outputs, controlling spend, and preserving the lessons that should make the next run better.
 
@@ -12,6 +14,16 @@ Rudder began as a fork of an early version of Paperclip. That gave the project a
 
 Rudder is built for the moment when agent work stops looking like a single prompt and starts looking like a real team.
 
+## A Better Harness for Agent Work
+
+We benchmarked Rudder, Codex CLI, and Claude Code on a random sample of professional tasks from OpenAI's GDPval dataset, with every system requesting the same model and reasoning effort. Rudder achieved the highest score under our local case-equal rubric metric.
+
+[![GDPval-based harness benchmark: Rudder 81.7, Codex CLI 75.7, and Claude Code 75.6.](docs/images/gdpval-harness-benchmark.png)](https://docs.rudderhq.dev/benchmarks/gdpval-harness)
+
+A model is only as effective as the environment around it. Rudder gives agents durable context, coordinated tools, structured work, review feedback, budgets, and memory—helping them stay focused, preserve what they learn, and produce stronger results across real work.
+
+The benchmark provides directional evidence that the harness and work environment matter; it is not an official GDPval score or leaderboard result. [Read the methodology and evidence limits.](https://docs.rudderhq.dev/benchmarks/gdpval-harness)
+
 ## The Work Loop
 
 Rudder is designed around the loop that makes agent work compound:
@@ -53,6 +65,14 @@ Rudder is open-source software for assigning, running, reviewing, and improving
 
 Rudder coordinates agents. It does not force one runtime, one model, one prompt format, or one execution environment.
 
+## Bring Your Own Runtime
+
+Rudder is the coordination layer, not the model provider. Run agents with the local tools and provider accounts you already use—including Codex, Claude Code, Cursor, OpenClaw, Bash, or your own HTTP service. Local runtimes keep their existing installation, login, and credentials; Rudder connects assignment, context, execution, review, budgets, and memory around them.
+
+![Rudder coordinates agent teams across local runtimes and provider environments.](docs/images/rudder-runtime-adapters.svg)
+
+[See supported runtimes and configuration options.](https://docs.rudderhq.dev/reference/runtime-types)
+
 ## Get Started
 
 ### Try Rudder

package/dist/index.js

@@ -27,6 +27,13 @@ var init_chat_inline_visuals = __esm({
   }
 });
 
+// ../packages/shared/dist/chat-subagents.js
+var init_chat_subagents = __esm({
+  "../packages/shared/dist/chat-subagents.js"() {
+    "use strict";
+  }
+});
+
 // ../packages/shared/dist/project-mentions.js
 var init_project_mentions = __esm({
   "../packages/shared/dist/project-mentions.js"() {
@@ -4728,6 +4735,7 @@ var init_dist = __esm({
     "use strict";
     init_browser_shortcuts();
     init_chat_inline_visuals();
+    init_chat_subagents();
     init_chat_work_manifest();
     init_constants();
     init_constants();