[Bug]: Codex cancels on code diff rejection

[vchernetskyi993]

Pre-submission checklist

• I have read the documentation
• I have updated the plugin and all dependencies to the latest versions
• I have searched for existing issues and discussions
• My issue is not a minor or cosmetic quirk (e.g., formatting, spacing, or other non-functional details)

Neovim version (nvim -v)

v0.12.0

Operating system/version

Ubuntu 24.04.4 LTS

Adapter and model

Codex and GPT-5.5

Describe the bug

Whenever I select Reject for code diff proposed by Codex ACP, I receive: You selected: Reject and the prompt immediately ends with ## Me ....

Also, I see a warning: [WARN] [acp:prompt_builder] Turn ended with stop_reason=cancelled.

Steps to reproduce

1. Open CodeCompanionChat.
2. Select Codex adapter.
3. Ask for any code change.
4. Reject.
5. The whole prompt processing is cancelled.

Expected behavior

When using Copilot (with any model) during reject I am receiving vim.ui.input query with title Rejection reason, and when I enter the reason it is acknowledged by the model with message User rejected the edits for ... with the reason ..., and execution continues.

Also, similar behavior can be observed inside codex CLI itself (i.e., it proposes changes, user rejects and writes a reason, then Codex applies rejection suggestion).

Screenshots or recordings (optional)

No response

minimal.lua file

---@diagnostic disable: missing-fields

vim.env.LAZY_STDPATH = ".repro"
load(vim.fn.system("curl -s https://raw.githubusercontent.com/folke/lazy.nvim/main/bootstrap.lua"))()

-- Your CodeCompanion setup
local plugins = {
	{
		"olimorris/codecompanion.nvim",
		dependencies = {
			{ "nvim-lua/plenary.nvim" },
			{
				"nvim-treesitter/nvim-treesitter",
				lazy = false,
				build = ":TSUpdate",
			},
		},
		opts = {
			adapters = {
				acp = {
					codex = function()
						return require("codecompanion.adapters").extend("codex", {
							defaults = {
								auth_method = "chatgpt",
							},
						})
					end,
				},
			},
			opts = {
				log_level = "DEBUG",
			},
		},
	},
}

require("lazy.minit").repro({ spec = plugins })

require("nvim-treesitter")
	.install({
		"lua",
		"markdown",
		"markdown_inline",
		"yaml",
	}, { summary = true, max_jobs = 10 })
	:wait(1800000)

Log output (optional)

[DEBUG] 2026-05-07 12:45:26
[acp] Process started: codex-acp
[DEBUG] 2026-05-07 12:45:26
[acp::connect_and_authenticate] Agent info: {
  agentCapabilities = {
    auth = {
      logout = vim.empty_dict()
    },
    loadSession = true,
    mcpCapabilities = {
      http = true,
      sse = false
    },
    promptCapabilities = {
      audio = false,
      embeddedContext = true,
      image = true
    },
    sessionCapabilities = {
      close = vim.empty_dict(),
      list = vim.empty_dict()
    }
  },
  agentInfo = {
    name = "codex-acp",
    title = "Codex",
    version = "0.13.0"
  },
  authMethods = { {
      description = "Use your ChatGPT login with Codex CLI (requires a paid ChatGPT subscription)",
      id = "chatgpt",
      name = "Login with ChatGPT"
    }, {
      description = "Requires setting the `CODEX_API_KEY` environment variable.",
      id = "codex-api-key",
      name = "Use CODEX_API_KEY",
      type = "env_var",
      vars = { {
          name = "CODEX_API_KEY"
        } }
    }, {
      description = "Requires setting the `OPENAI_API_KEY` environment variable.",
      id = "openai-api-key",
      name = "Use OPENAI_API_KEY",
      type = "env_var",
      vars = { {
          name = "OPENAI_API_KEY"
        } }
    } },
  protocolVersion = 1
}
[DEBUG] 2026-05-07 12:45:26
[acp] Initialized (protocol_version=1)
[DEBUG] 2026-05-07 12:45:26
[acp] Authenticated
[DEBUG] 2026-05-07 12:45:27
[acp] Config options: { {
    category = "model",
    currentValue = "gpt-5.5",
    description = "Choose which model Codex should use",
    id = "model",
    name = "Model",
    options = { {
        description = "Frontier model for complex coding, research, and real-world work.",
        name = "GPT-5.5",
        value = "gpt-5.5"
      }, {
        description = "Strong model for everyday coding.",
        name = "gpt-5.4",
        value = "gpt-5.4"
      }, {
        description = "Small, fast, and cost-efficient model for simpler coding tasks.",
        name = "GPT-5.4-Mini",
        value = "gpt-5.4-mini"
      }, {
        description = "Coding-optimized model.",
        name = "gpt-5.3-codex",
        value = "gpt-5.3-codex"
      }, {
        description = "Ultra-fast coding model (preview).",
        name = "gpt-5.3-codex-spark-preview",
        value = "gpt-5.3-codex-spark-preview"
      }, {
        description = "Optimized for professional work and long-running agents.",
        name = "gpt-5.2",
        value = "gpt-5.2"
      } },
    type = "select"
  }, {
    category = "thought_level",
    currentValue = "xhigh",
    description = "Choose how much reasoning effort the model should use",
    id = "reasoning_effort",
    name = "Reasoning Effort",
    options = { {
        description = "Fast responses with lighter reasoning",
        name = "Low",
        value = "low"
      }, {
        description = "Balances speed and reasoning depth for everyday tasks",
        name = "Medium",
        value = "medium"
      }, {
        description = "Greater reasoning depth for complex problems",
        name = "High",
        value = "high"
      }, {
        description = "Extra high reasoning depth for complex problems",
        name = "Xhigh",
        value = "xhigh"
      } },
    type = "select"
  } }
[DEBUG] 2026-05-07 12:45:27
[acp::_establish_session] New session config options applied
[DEBUG] 2026-05-07 12:45:27
[acp] Session established: 019e01d3-f17b-7a30-863e-c4d8097a1382
[DEBUG] 2026-05-07 12:45:27
[acp::commands] Registered 7 commands for session 019e01d3-f17b-7a30-863e-c4d8097a1382
[DEBUG] 2026-05-07 12:45:28
[acp] Config options: { {
    category = "model",
    currentValue = "gpt-5.5",
    description = "Choose which model Codex should use",
    id = "model",
    name = "Model",
    options = { {
        description = "Frontier model for complex coding, research, and real-world work.",
        name = "GPT-5.5",
        value = "gpt-5.5"
      }, {
        description = "Strong model for everyday coding.",
        name = "gpt-5.4",
        value = "gpt-5.4"
      }, {
        description = "Small, fast, and cost-efficient model for simpler coding tasks.",
        name = "GPT-5.4-Mini",
        value = "gpt-5.4-mini"
      }, {
        description = "Coding-optimized model.",
        name = "gpt-5.3-codex",
        value = "gpt-5.3-codex"
      }, {
        description = "Ultra-fast coding model (preview).",
        name = "gpt-5.3-codex-spark-preview",
        value = "gpt-5.3-codex-spark-preview"
      }, {
        description = "Optimized for professional work and long-running agents.",
        name = "gpt-5.2",
        value = "gpt-5.2"
      } },
    type = "select"
  }, {
    category = "thought_level",
    currentValue = "xhigh",
    description = "Choose how much reasoning effort the model should use",
    id = "reasoning_effort",
    name = "Reasoning Effort",
    options = { {
        description = "Fast responses with lighter reasoning",
        name = "Low",
        value = "low"
      }, {
        description = "Balances speed and reasoning depth for everyday tasks",
        name = "Medium",
        value = "medium"
      }, {
        description = "Greater reasoning depth for complex problems",
        name = "High",
        value = "high"
      }, {
        description = "Extra high reasoning depth for complex problems",
        name = "Xhigh",
        value = "xhigh"
      } },
    type = "select"
  } }
[DEBUG] 2026-05-07 12:45:28
[acp::set_config_option] Changed model to gpt-5.5
[INFO] 2026-05-07 12:45:45
Chat request started
[DEBUG] 2026-05-07 12:45:52
[ACP::Handler] Processing tool call {
  kind = "search",
  rawInput = {
    call_id = "call_lJu6nBoUzDtsLaAVGdbPDSiY",
    command = { "/bin/bash", "-lc", "rg --files -g '!*__pycache__*' | head -50" },
    cwd = "/home/chernetskyi/repo/personal/repros/codecompanion-codex-reject",
    parsed_cmd = { {
        cmd = "rg --files -g '!*__pycache__*'",
        path = vim.NIL,
        type = "list_files"
      } },
    process_id = "82706",
    source = "unified_exec_startup",
    turn_id = "019e01d4-3559-73f1-966c-697d5a729713"
  },
  sessionUpdate = "tool_call",
  status = "in_progress",
  title = "List /home/chernetskyi/repo/personal/repros/codecompanion-codex-reject",
  toolCallId = "call_lJu6nBoUzDtsLaAVGdbPDSiY"
}
[DEBUG] 2026-05-07 12:45:52
[ACP::Handler] Processing tool call {
  rawOutput = {
    aggregated_output = "minimal.lua\n",
    call_id = "call_lJu6nBoUzDtsLaAVGdbPDSiY",
    command = { "/bin/bash", "-lc", "rg --files -g '!*__pycache__*' | head -50" },
    cwd = "/home/chernetskyi/repo/personal/repros/codecompanion-codex-reject",
    duration = {
      nanos = 15271,
      secs = 0
    },
    exit_code = 0,
    formatted_output = "minimal.lua\n",
    parsed_cmd = { {
        cmd = "rg --files -g '!*__pycache__*'",
        path = vim.NIL,
        type = "list_files"
      } },
    process_id = "82706",
    source = "unified_exec_startup",
    status = "completed",
    stderr = "",
    stdout = "minimal.lua\n",
    turn_id = "019e01d4-3559-73f1-966c-697d5a729713"
  },
  sessionUpdate = "tool_call_update",
  status = "completed",
  toolCallId = "call_lJu6nBoUzDtsLaAVGdbPDSiY"
}
[DEBUG] 2026-05-07 12:45:54
[ACP::Handler] Processing tool call {
  content = { {
      newText = 'def main():\n    print("Hello, world!")\n\n\nif __name__ == "__main__":\n    main()\n',
      path = "/home/chernetskyi/repo/personal/repros/codecompanion-codex-reject/hello.py",
      type = "diff"
    } },
  kind = "edit",
  locations = { {
      path = "/home/chernetskyi/repo/personal/repros/codecompanion-codex-reject/hello.py"
    } },
  rawInput = {
    auto_approved = true,
    call_id = "call_UBRVsyugnnOB9EC7rXsFq9Dw",
    changes = {
      ["/home/chernetskyi/repo/personal/repros/codecompanion-codex-reject/hello.py"] = {
        content = 'def main():\n    print("Hello, world!")\n\n\nif __name__ == "__main__":\n    main()\n',
        type = "add"
      }
    },
    turn_id = "019e01d4-3559-73f1-966c-697d5a729713"
  },
  sessionUpdate = "tool_call",
  status = "in_progress",
  title = "Edit /home/chernetskyi/repo/personal/repros/codecompanion-codex-reject/hello.py",
  toolCallId = "call_UBRVsyugnnOB9EC7rXsFq9Dw"
}
[DEBUG] 2026-05-07 12:45:54
[ACP::Handler] Permission Request
  id: call_UBRVsyugnnOB9EC7rXsFq9Dw
  kind: edit
  Edit /home/chernetskyi/repo/personal/repros/codecompanion-codex-reject/hello.py
  options: { "allow_once", "reject_once" }
[DEBUG] 2026-05-07 12:45:59
[approval_prompt] User selected: View
[DEBUG] 2026-05-07 12:45:59
[acp::request_permission] Opening diff for review
[DEBUG] 2026-05-07 12:46:03
[acp::request_permission] User selected option: reject_once (abort)
[WARN] 2026-05-07 12:46:05
[acp::prompt_builder] Turn ended with stop_reason=cancelled
[INFO] 2026-05-07 12:46:05
Chat request finished

Minimal reproduction confirmation

• Yes, I have tested and provided a minimal.lua file that reproduces the issue

[olimorris]

N.B. This response has been generated by the CodeCompanion Issues Agent

Hi @vchernetskyi993, thanks for the detailed bug report — the description is clear and the contrast with Copilot's behaviour is very helpful for understanding the expected behaviour. However, there are a few issues with the minimal.lua that prevent the maintainer from being able to reproduce this:

• Missing core dependencies: nvim-treesitter is a required dependency of CodeCompanion and must be included in the minimal.lua. Please use the template provided in the contributing guide as your base.
• External ACP binary not addressed: The Codex adapter requires an external CLI tool (the Zed ACP adapter for Codex) to be installed and available on your PATH. Please confirm this is set up in your reproduction environment and add a note to your issue.
• No log output provided: Since this involves ACP-level communication and a stop_reason=cancelled warning, the debug log would be invaluable. Please enable debug logging (log_level = "DEBUG" in your opts.opts — see configuration docs) and share the relevant portion of the log file with your next update.

[vchernetskyi993]

Missing core dependencies: nvim-treesitter is a required dependency of CodeCompanion and must be included in the minimal.lua. Please use the template provided in the contributing guide as your base.

I thought that it's irrelevant for the issue at hand, thus removed it initially. Reverted and re-tested with treesitter in place.

External ACP binary not addressed: The Codex adapter requires an external CLI tool (the Zed ACP adapter for Codex) to be installed and available on your PATH. Please confirm this is set up in your reproduction environment and add a note to your issue.

$ which codex-acp
/home/chernetskyi/.nvm/versions/node/v24.14.1/bin/codex-acp

No log output provided: Since this involves ACP-level communication and a stop_reason=cancelled warning, the debug log would be invaluable. Please enable debug logging (log_level = "DEBUG" in your opts.opts — see configuration docs) and share the relevant portion of the log file with your next update.

I'm sorry, accidentally dropped log_level option, while was copying my adapter configuration. Reverted and added logs to description.

[olimorris]

I don't see how this is a CodeCompanion issue. If CodeCompanion is sending the rejection and Codex is choosing to cancel, then CodeCompanion is correctly handling that.

[vchernetskyi993]

I don't see how this is a CodeCompanion issue. If CodeCompanion is sending the rejection and Codex is choosing to cancel, then CodeCompanion is correctly handling that.

Okay, thanks and sorry to spend your time on this. Tried to log an issue in the codex-acp, maybe folks there would have some pointers.

The issue is quite annoying for me, as I regularly like to refine code produced by LLM, and in this case the whole flow is aborted and needs restart even for small corrections.

[olimorris]

One aspect of the Agent Client Protocol that I would really like is the ability to reject and provide feedback. I don't believe that's in the spec but being able to say why we're rejecting it would be really clean.