AgentArmorHQ
Sign in
  • AgentArmor teamTeam3d agoNeeds helpSolved

    Continue 1.0.9 shows reasoning after tool calls instead of collapsing it

    In Continue 1.0.9 on Linux with VS Code 1.99.3, Qwen3 reasoning output is collapsed before a tool call, but any follow-up reasoning after the tool call is shown to the user as plain text, including after consecutive tool calls. Is there a config option to collapse this follow-up reasoning, or is this a known bug?

    Originally asked on GitHub: https://github.com/continuedev/continue/issues/5684

    4 answers

    • Best answerAgentArmor teamTeam3d ago

      This issue was closed as NOT_PLANNED in September 2025 after going stale, and the thread contains no fix or workaround. A maintainer asked the reporter to use the message toolbar Copy button to capture the full assistant turn and determine whether the model was emitting thinking tags that Continue fails to parse. The reporter replied that Copy only grabs the segment after the last tool call, not the whole turn. In a follow-up test with two consecutive directory_get_employees tool calls, the pasted output still showed the model's thinking reasoning block as plain visible text, so the entire output was rendered to the user instead of being collapsed. No config option that controls this behavior was identified in the thread. General advice: since the thread found no workaround, consider disabling reasoning output through the model endpoint or switching to a non-reasoning model until this is addressed. Security note: the configuration in the report contains a model API key and a GitHub personal access token; keep those in environment variables instead of the config file.

      Did this work for you?
    • AgentArmor teamTeam3d ago

      What a GitHub user tried on GitHub (Outcome unclear)

      Reproduced the issue with a Qwen3 reasoning model and the GitHub MCP server, prompting for two consecutive tool calls (get_me and search_repositories). The initial reasoning collapsed, but all follow-up reasoning was shown as user-facing text.

      Source: https://github.com/continuedev/continue/issues/5684 (opens external site)

      Did this work for you?
    • AgentArmor teamTeam3d ago

      What @nejch tried on GitHub (Didn't work)

      Tried the message toolbar's Copy button to capture the full assistant turn for debugging. It only copied the last portion after the final tool call, not the entire process.

      Source: https://github.com/continuedev/continue/issues/5684#issuecomment-2914207441 (opens external site)

      Did this work for you?
    • AgentArmor teamTeam3d ago

      What @nejch tried on GitHub (Outcome unclear)

      Ran a separate test with two consecutive directory_get_employees tool calls. The pasted output still showed the thinking block, and the entire output was rendered to the user.

      Source: https://github.com/continuedev/continue/issues/5684#issuecomment-2914207441 (opens external site)

      Did this work for you?

    Add your answer or comment