Skip to content
DevAgentRadarDevAgentRadar
Sign in
Proposals/Pin Cline models so v4.1.21 does not default Git
proposaleveryoneP4Worth a lookCline

Pin Cline models so v4.1.21 does not default GitHub Copilot and Vertex to Claude Opus 5.5

Cline v4.1.21 changes the resolved default for 19 unpinned providers, sending 11 of them (including GitHub Copilot and Vertex) to Claude Opus 5.5, adds the Japanese open-weight ai& provider, and parses rule/skill frontmatter with js-yaml 4.3.2. Pin models on upgrade; local long replies now compact and retry once.

Why this loop

Unpinned Cline providers no longer keep a stable default. After v4.1.21, 19 providers that do not pin a model get a new resolved default, and 11 of those—including GitHub Copilot and Vertex—now land on Claude Opus 5.5. That can change cost, quality, and coding style with no explicit settings edit. Pin an explicit model ID on every provider you actually use before or immediately after upgrading. Take the upgrade anyway: js-yaml 4.3.2 fixes the parser that reads rule and skill frontmatter; llama.cpp, Ollama, and LM Studio no longer end the task when a long reply hits leftover context (they compact once, then concise-retry, and keep the partial answer—those servers still ignore your output budget); failed tasks reopen as errors with retry instead of looking completed; empty commands no longer dump a raw JSON blob; Windows @-mentions show real nested paths. If you want Japanese open-weight models, add the new OpenAI-compatible ai& provider and pin a model there too.

Proposed actions

  1. Upgrade Cline to v4.1.21 so rule and skill frontmatter is parsed with js-yaml 4.3.2, failed tasks reopen with a retry option instead of looking completed, empty commands no longer print a raw JSON blob like [{"query":"git add -A","result":"","success":true}], and canceling during an empty-response retry takes effect immediately.
  2. After the upgrade, pin an explicit model ID on every Cline provider you use that does not already pin one—especially GitHub Copilot and Vertex—because v4.1.21 now changes the resolved default for 19 unpinned providers, 11 of them to Claude Opus 5.5.
  3. If you use llama.cpp, Ollama, or LM Studio in Cline, do not raise the output-token budget expecting those servers to honor it: they still cap generation at leftover context. Rely on v4.1.21 compact-once-and-retry for long replies; if compaction cannot help, concise-retry still runs and the partial answer is kept.
  4. To use Japanese open-weight models, add Cline’s new ai& provider as an OpenAI-compatible endpoint and pin a specific model from the refreshed 6,386-model catalog (209 providers) instead of leaving the provider unpinned.
  5. On Windows Cline v4.1.21, @-mention nested files by their real relative paths (a subfolder README is no longer shown as /README.md, and the same open file no longer appears twice).

Agent prompt

Paste into your agent or query via MCP (get_agent_prompt) — free, no extra AI cost

Cline / .clinerules task

DevAgentRadar → Cline

You are helping me adopt a real coding-assistant change. Work only from the facts below. Do not invent features.

Context

Assistant: Cline Proposal: Pin Cline models so v4.1.21 does not default GitHub Copilot and Vertex to Claude Opus 5.5 Summary: Cline v4.1.21 changes the resolved default for 19 unpinned providers, sending 11 of them (including GitHub Copilot and Vertex) to Claude Opus 5.5, adds the Japanese open-weight ai& provider, and parses rule/skill frontmatter with js-yaml 4.3.2. Pin models on upgrade; local long replies now compact and retry once. Primary source: https://github.com/cline/cline/releases/tag/v4.1.21

Why it matters

Unpinned Cline providers no longer keep a stable default. After v4.1.21, 19 providers that do not pin a model get a new resolved default, and 11 of those—including GitHub Copilot and Vertex—now land on Claude Opus 5.5. That can change cost, quality, and coding style with no explicit settings edit. Pin an explicit model ID on every provider you actually use before or immediately after upgrading. Take the upgrade anyway: js-yaml 4.3.2 fixes the parser that reads rule and skill frontmatter; llama.cpp, Ollama, and LM Studio no longer end the task when a long reply hits leftover context (they compact once, then concise-retry, and keep the partial answer—those servers still ignore your output budget); failed tasks reopen as errors with retry instead of looking completed; empty commands no longer dump a raw JSON blob; Windows @-mentions show real nested paths. If you want Japanese open-weight models, add the new OpenAI-compatible ai& provider and pin a model there too.

Suggested actions

  1. Upgrade Cline to v4.1.21 so rule and skill frontmatter is parsed with js-yaml 4.3.2, failed tasks reopen with a retry option instead of looking completed, empty commands no longer print a raw JSON blob like [{"query":"git add -A","result":"","success":true}], and canceling during an empty-response retry takes effect immediately.
  2. After the upgrade, pin an explicit model ID on every Cline provider you use that does not already pin one—especially GitHub Copilot and Vertex—because v4.1.21 now changes the resolved default for 19 unpinned providers, 11 of them to Claude Opus 5.5.
  3. If you use llama.cpp, Ollama, or LM Studio in Cline, do not raise the output-token budget expecting those servers to honor it: they still cap generation at leftover context. Rely on v4.1.21 compact-once-and-retry for long replies; if compaction cannot help, concise-retry still runs and the partial answer is kept.
  4. To use Japanese open-weight models, add Cline’s new ai& provider as an OpenAI-compatible endpoint and pin a specific model from the refreshed 6,386-model catalog (209 providers) instead of leaving the provider unpinned.
  5. On Windows Cline v4.1.21, @-mention nested files by their real relative paths (a subfolder README is no longer shown as /README.md, and the same open file no longer appears twice).

After you finish

Do not report this as applied to DevAgentRadar. You cannot write the visitor's loop.

Tell the human: open https://devagentradar.com/proposals/cline-v4-1-21-pin-cline-models-so-v4-1-21-does-not-default-github-copilo and mark Applied, Skipped, or Failed. Proposal id: 9d2d00f6-2b62-488f-87bb-d7ae8f144e75

Your job

  1. Restate the change in one sentence.
  2. Propose a minimal plan for my repo (or a throwaway pilot).
  3. Implement only what I approve; prefer small diffs and tests.
  4. Call out risks (permissions, breaking APIs, cost).

Start by confirming you understood the proposal.

modelsecurityRelease source ↗

Your loop

This browser · no sign-in · not shared as “you”

After you run the prompt

Only you can mark this. Agents cannot write your loop.

Your decision stays on this device. A public tally appears after a few votes.

Originating release signal

Clinev4.1.21Sep 24, 2026

v4.1.21

New provider: ai&, an OpenAI-compatible endpoint serving open-weight models from Japan. · Refreshed the model catalog to 6,386 models across 209 providers. · The resolved default model changes for 19 providers that do not pin one, 11 of them to Claude Opus 5.5 (including GitHub Copilot and Vertex). · If you use one of those providers without pinning a model, expect a different default. · Raised the minimum js-yaml version to 4.3.2 to pick up a security fix in the parser used to read rule and skill frontmatter. · +8 more changes
Verified excerpt — the source's own words

Added

  • New provider: ai&, an OpenAI-compatible endpoint serving open-weight models from Japan.

Changed

  • Refreshed the model catalog to 6,386 models across 209 providers. The resolved default model changes for 19 providers that do not pin one, 11 of them to Claude Opus 5.5 (including GitHub Copilot and Vertex). If you use one of those providers without pinning a model, expect a different default.
  • Raised the minimum js-yaml version to 4.3.2 to pick up a security fix in the parser used to read rule and skill frontmatter.

Fixed

  • Long replies on local models (llama.cpp, Ollama, LM Studio) that hit the output-token limit now compact the conversation and retry once instead of ending the task. These servers cap generation at whatever context is left, regardless of the output budget you set. If compaction cannot help, the existing concise-retry recovery still runs, and the partial answer is kept.
  • Reopening a task that failed now shows the error with the retry option, instead of presenting it as a completed task.
  • A command that prints nothing no longer shows a raw JSON blob like [{"query":"git add -A","result":"","success":true}] as its output.

Excerpt ends here — this release continues at the source ↗.

Primary source ↗