Upgrade Gemini CLI to v0.56.0-preview.1 to end false capacity exhaustion
Move to Gemini CLI v0.56.0-preview.1 so false capacity-exhaustion failures stop, real capacity exhaustion is terminal, MCP OAuth refreshes with the stored client ID, and the new local eval report path is available.Why this loop
v0.56.0-preview.1 includes fix(core,cli) that stops false model capacity exhaustion and corrects core quota-lookup model mapping, plus reclassifying real capacity exhaustion as a terminal error. Staying on 0.54.x or 0.55.0-preview.1 leaves agents retrying phantom quota failures or continuing after a true capacity stop. The same tag’s fix(core) refreshes MCP OAuth tokens with the stored client ID, which prevents refresh failures from a mismatched client id. feat(evals) adds a local report command and developer documentation so eval reports can be generated without a remote runner.
Proposed actions
- Upgrade the CLI pin to exactly this tag: npm install -g @google/gemini-cli@0.56.0-preview.1 && gemini --version — fail the change if the printed version is not 0.56.0-preview.1.
- In package.json, lockfiles, CI install steps, and devcontainer/Docker images, replace every @google/gemini-cli version below 0.56.0-preview.1 (including 0.54.x and 0.55.0-preview.1) with @google/gemini-cli@0.56.0-preview.1.
- In agent retry wrappers and runbooks, treat remaining capacity-exhaustion errors as terminal after this upgrade: delete backoff/retry on that signal; only the false-positive quota-lookup mapping was fixed.
- After the upgrade, reconnect every MCP server whose OAuth tokens were issued by the previous CLI so refresh uses the stored client ID from fix(core), then confirm a tool call succeeds without a client-id mismatch on refresh.
- Switch eval workflows to the local report command added in feat(evals) in v0.56.0-preview.1; generate the report locally using this release’s developer documentation and keep the artifact next to the eval fixtures.
Agent prompt
Paste into your agent or query via MCP (get_agent_prompt) — free, no extra AI costGoogle Antigravity / agent task
DevAgentRadar → Google Antigravity
Goal: turn this release signal into a safe pilot plan for my stack.
Context
Assistant: Gemini CLI Proposal: Upgrade Gemini CLI to v0.56.0-preview.1 to end false capacity exhaustion Summary: Move to Gemini CLI v0.56.0-preview.1 so false capacity-exhaustion failures stop, real capacity exhaustion is terminal, MCP OAuth refreshes with the stored client ID, and the new local eval report path is available. Primary source: https://github.com/google-gemini/gemini-cli/releases/tag/v0.56.0-preview.1
Why it matters
v0.56.0-preview.1 includes fix(core,cli) that stops false model capacity exhaustion and corrects core quota-lookup model mapping, plus reclassifying real capacity exhaustion as a terminal error. Staying on 0.54.x or 0.55.0-preview.1 leaves agents retrying phantom quota failures or continuing after a true capacity stop. The same tag’s fix(core) refreshes MCP OAuth tokens with the stored client ID, which prevents refresh failures from a mismatched client id. feat(evals) adds a local report command and developer documentation so eval reports can be generated without a remote runner.
Suggested actions
- Upgrade the CLI pin to exactly this tag: npm install -g @google/gemini-cli@0.56.0-preview.1 && gemini --version — fail the change if the printed version is not 0.56.0-preview.1.
- In package.json, lockfiles, CI install steps, and devcontainer/Docker images, replace every @google/gemini-cli version below 0.56.0-preview.1 (including 0.54.x and 0.55.0-preview.1) with @google/gemini-cli@0.56.0-preview.1.
- In agent retry wrappers and runbooks, treat remaining capacity-exhaustion errors as terminal after this upgrade: delete backoff/retry on that signal; only the false-positive quota-lookup mapping was fixed.
- After the upgrade, reconnect every MCP server whose OAuth tokens were issued by the previous CLI so refresh uses the stored client ID from fix(core), then confirm a tool call succeeds without a client-id mismatch on refresh.
- Switch eval workflows to the local report command added in feat(evals) in v0.56.0-preview.1; generate the report locally using this release’s developer documentation and keep the artifact next to the eval fixtures.
Config surfaces this release may change
- MCP servers — check your repo before applying
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/gemini-cli-v0-56-0-preview-1-upgrade-gemini-cli-to-v0-56-0-preview-1-to- and mark Applied, Skipped, or Failed. Proposal id: 16c47249-8b9f-47dd-9e25-a503193e16eb
Please:
- Map the change to concrete pilot steps
- Flag security / permission implications
- Keep the pilot reversible
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.