# Difference Between `--resume` and `--fresh` Flags in Codex: Routing Controls Explained > Understand the difference between Codex --resume and --fresh flags. --resume continues your last session, while --fresh starts a new one. Learn routing controls for openai/codex-plugin-cc. - Repository: [OpenAI/codex-plugin-cc](https://github.com/openai/codex-plugin-cc) - Tags: deep-dive - Published: 2026-07-10 --- **The `--resume` flag forces the Codex runtime to continue the most recent rescue thread by injecting `--resume-last` into the task command, while `--fresh` explicitly starts a brand-new task session that ignores any prior work.** The `openai/codex-plugin-cc` repository provides these routing controls to help developers manage how new tasks relate to previous Codex sessions. Understanding the difference between `--resume` and `--fresh` flags in Codex is essential for controlling whether your investigation builds upon existing rescue work or begins as an independent analysis. ## How `--resume` and `--fresh` Work Under the Hood Both flags operate as routing controls in the **Codex CLI Runtime skill**, determining how the plugin interprets your intent regarding prior repository state. ### The `--resume` Flag When you provide `--resume`, the plugin strips this flag from your input and automatically **adds `--resume-last`** to the underlying `task` invocation. This forces the runtime to continue the most recent rescue thread for the repository, regardless of whether your prompt wording seems ambiguous. According to the implementation in [`plugins/codex/skills/codex-cli-runtime/SKILL.md`](https://github.com/openai/codex-plugin-cc/blob/main/plugins/codex/skills/codex-cli-runtime/SKILL.md) (lines 31-34), this mode tells the system to "always use `task --resume-last`, even if the request text is ambiguous." ### The `--fresh` Flag Conversely, `--fresh` strips the flag from your task text but **does not add `--resume-last`** to the command. This ensures the runtime starts a **brand-new task**, creating a completely independent investigation that ignores any existing rescue threads. The same source file specifies this behavior as "always use a fresh `task` run, even if the request sounds like a follow-up." ## When to Use Each Flag Choosing the correct flag depends on whether you want to build upon previous Codex work or start clean. - **Use `--resume`** when you explicitly want to continue working, apply the top fix from a previous run, or build upon the last session's context. This maintains continuity with the existing rescue thread. - **Use `--fresh`** when you need a new, independent investigation or fix, even if your natural language sounds like a follow-up question. This prevents the runtime from inheriting context from previous sessions. If you omit both flags, the plugin will offer to continue the latest rescue thread based on its interpretation of your request, as documented in the repository's [`README.md`](https://github.com/openai/codex-plugin-cc/blob/main/README.md) (lines 40-48). ## Implementation Details in the Source Code The flag handling logic resides in the **Codex CLI Runtime skill** at [`plugins/codex/skills/codex-cli-runtime/SKILL.md`](https://github.com/openai/codex-plugin-cc/blob/main/plugins/codex/skills/codex-cli-runtime/SKILL.md). This file defines the transformation rules that convert user-facing commands into internal `task` invocations. The user-facing documentation in [`README.md`](https://github.com/openai/codex-plugin-cc/blob/main/README.md) reinforces that these flags provide explicit control over the default behavior, while [`plugins/codex/commands/rescue.md`](https://github.com/openai/codex-plugin-cc/blob/main/plugins/codex/commands/rescue.md) documents how end-users access these options via the `/codex:rescue` command. ## Code Examples The following examples demonstrate how these flags transform into internal commands: ```bash # Continue the most recent rescue thread (resume mode) /codex:rescue --resume apply the top fix from the last run # → Internally becomes: task --resume-last "apply the top fix from the last run" # Start a brand-new task, ignoring any prior rescue work (fresh mode) /codex:rescue --fresh investigate why the tests are flaky # → Internally becomes: task "investigate why the tests are flaky" ``` Notice that `--resume` injects the `--resume-last` parameter into the underlying `task` command, while `--fresh` results in a clean `task` invocation without resume parameters. ## Summary - **`--resume`** strips the flag and adds `--resume-last` to continue the most recent rescue thread, ideal for building upon previous work. - **`--fresh`** strips the flag and starts a new task without resume parameters, perfect for independent investigations. - Both behaviors are defined in [`plugins/codex/skills/codex-cli-runtime/SKILL.md`](https://github.com/openai/codex-plugin-cc/blob/main/plugins/codex/skills/codex-cli-runtime/SKILL.md) and exposed to users via the `/codex:rescue` command. - Omitting both flags leaves the decision to the plugin's default interpretation logic. ## Frequently Asked Questions ### What happens if I use both `--resume` and `--fresh` flags together? The source code in [`plugins/codex/skills/codex-cli-runtime/SKILL.md`](https://github.com/openai/codex-plugin-cc/blob/main/plugins/codex/skills/codex-cli-runtime/SKILL.md) does not explicitly define this conflict scenario. According to the implementation logic, each flag serves an opposite purpose—one forces continuation while the other forces a fresh start. You should specify only one flag to ensure predictable behavior, as the plugin expects mutually exclusive routing directives. ### Can I resume a specific rescue thread instead of just the last one? The current implementation only supports `--resume-last` as the underlying mechanism when using `--resume`. The flag is designed to continue the most recent rescue thread specifically, not arbitrary historical threads. For targeted recovery of specific sessions, you would need to use the underlying `task` command directly with appropriate identifiers. ### Why does the plugin strip these flags from the task text? The flags are stripped because they are **routing controls** meant for the plugin's command parser, not for the underlying task execution. The plugin consumes these flags to determine the execution mode, then passes only the remaining task description to the runtime. This ensures the actual task content remains clean and focused on the work to be performed. ### Where can I find the official documentation for these flags? The authoritative documentation resides in three locations within the `openai/codex-plugin-cc` repository: the implementation details in [`plugins/codex/skills/codex-cli-runtime/SKILL.md`](https://github.com/openai/codex-plugin-cc/blob/main/plugins/codex/skills/codex-cli-runtime/SKILL.md) (lines 31-34), user-facing descriptions in [`README.md`](https://github.com/openai/codex-plugin-cc/blob/main/README.md) (lines 40-48), and command syntax in [`plugins/codex/commands/rescue.md`](https://github.com/openai/codex-plugin-cc/blob/main/plugins/codex/commands/rescue.md).