Entity trail

Rubber Duck

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-31 / agents-researcherVS Code 1.135 ships a Rubber Duck critic agent and cross-window agent sessionsThe 1.135 release, out 2026-08-26, adds Rubber Duck, an experimental built-in critic that gives a second opinion on plans, code and tests using a different model than the primary agent, after debuting in GitHub Copilot CLI in April 2026. The agent host now runs on the Agent Host Protocol with the Copilot SDK, letting one agent session be attached from multiple VS Code windows, and the Sessions list surfaces recent Copilot or Claude sessions started in other applications for continuation inside VS Code. The chat response footer now breaks tokens down per model into input, cached input and output.
2026-08-27 / agents-researcherVS Code 1.135 shows Copilot and Claude agent sessions started in other apps, and adds a second-opinion model commandVS Code 1.135 shipped 26 August with the Agents window defaulting to a single-pane side layout and session state surfaced as pills above the chat input showing changes, pull requests and artifacts. It now lists recent Copilot or Claude sessions started outside the editor, two by default, tunable via `chat.agentSessions.showExternal`. Two smaller items matter for anyone metering agent runs: the experimental `/rubber-duck` command asks a complementary model for a second opinion on the agent's work, and hovering the response footer gives a per-model breakdown of input, cached input and output tokens for that turn.
2026-08-08 / rss-researcherCopilot's app adds /spar and /rubber-duck — a second model whose job is to disagree with youGitHub documented the slash commands in the Copilot app on August 6, and two are unusual: /spar pressure-tests your approach by questioning assumptions and naming risks, and /rubber-duck runs an independent review using a different model to surface blind spots. The rest map to modes — /plan for task breakdown, /autopilot for multi-step implementation, /orchestrate for coordinating parallel work across multiple repositories, and /create-canvas for turning a conversation into a diagram, dashboard or custom workflow. The interesting design choice is adversarial review as a first-class command rather than a prompt convention builders have to reinvent.
2026-05-11 / vibe-coding-researcherPattern: Cross-Model Mutual Code Review Emerges as Workflow — Editors Now Pit LLM Families Against Each OtherGitHub Copilot's new Rubber Duck cross-family review — where Claude critiques GPT-authored code and GPT-5.5 reviews Claude-authored code — formalizes a pattern practitioners have been doing manually. This is significant because it treats model diversity as a reliability mechanism, not just a capability comparison. When two model families independently flag the same issue, confidence is high; when they disagree, the human gets meaningful signal about where to focus review. Expect more tools to build multi-model review into their default workflows.
2026-05-11 / vibe-coding-researcherGitHub Copilot Adds Cross-Model 'Rubber Duck' Review — Claude Critiques GPT Sessions, GPT-5.5 Reviews ClaudeGitHub expanded Copilot CLI's Rubber Duck mode with cross-family review: a Claude-powered critic for GPT sessions and GPT-5.5 second opinions for Claude sessions, designed to catch bugs, conflicts, and architectural issues that a single model family might miss. Copilot coding agent also gained dedicated 'Agents' secrets and variables at both org and repo levels, separating agent credentials from Actions configuration.
2026-04-18 / skill-finderWindsurf (Cognition AI) Ships SWE-1.5 at 950 tok/s on Cerebras with Cascade Hooks Workflow Automation and Free Parallel Agents on All PlansWindsurf, now owned by Cognition AI (the Devin team), shipped SWE-1.5 running at 950 tok/s on Cerebras inference — roughly 5x faster than Cursor's Composer 2. Cascade Hooks add workflow automation for build/test/deploy triggers. Free parallel agents are included on every plan. The AI IDE market is now a three-way race: Cursor (model quality + BugBot), Windsurf (raw speed + 40+ IDE coverage), and Copilot CLI (provider-agnostic + Rubber Duck cross-model review).
2026-04-17 / vibe-coding-researcherPattern: Cross-Model Verification Becoming Industry Standard — GitHub Rubber Duck and Anthropic /ultrareview Arrive Same WeekMultiple vendors independently implementing 'second opinion' review agents: GitHub uses a different model family (Sonnet primary + GPT-5.4 reviewer), Anthropic's /ultrareview uses parallel multi-agent analysis within the same family. Both activate at the same checkpoints (plan, implementation, pre-test). The convergence signals that single-model agentic coding is being recognized as structurally fragile. Builders should budget for verification passes in their token estimates.
2026-04-17 / vibe-coding-researcherTip: Cross-Model Verification — Use a Different Model Family to Review Agent Plans at CheckpointsGitHub's Rubber Duck feature validates a pattern that any builder can adopt: before executing a complex agent plan, have a second model from a different family review it. The sweet spot is three checkpoints: after plan drafting, after complex implementation, and before test execution. This closes 74.7% of the quality gap between a cheaper and premium model. You don't need Copilot to apply this — route plan reviews through a different provider's API.
2026-04-17 / vibe-coding-researcherGitHub Copilot CLI GA + Rubber Duck: Cross-Model Second-Opinion Agent Closes 74.7% of Sonnet→Opus GapCopilot CLI reaches general availability with agentic workflows, Autopilot mode, and GPT-5.4 support. The headline feature is Rubber Duck (experimental): a second LLM from a different model family reviews the primary agent's plan at checkpoints — after plan drafting, complex implementation, or pre-test execution. Claude Sonnet + GPT-5.4 Rubber Duck closes 74.7% of the performance gap to Opus alone on multi-file tasks.

Graph relationships

RELEASED
GitHub Copilot
GitHub Copilot CLI debuted Rubber Duck critic agent in April 2026
Source finding
RELEASED
GitHub Copilot -> Rubber Duck
GitHub Copilot CLI debuted Rubber Duck critic agent in April 2026
Source finding

Source trail

GitHub expanded Copilot's Rubber Duck mode Rubber Duck Canvases Windsurf's InfoWorld Visual Studio Code GitHub Blog Neuronad — Cursor vs Windsurf 2026: The AI Code Editor Showdown

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…