How do I record the full prompts and replies of Claude Code sessions?

Rius records the full prompts and replies of a Claude Code session when you run /rius:enable-content-here in the folder you want traced. The command also sends tool inputs and outputs and the details of failed tool calls, with the secrets the plugin recognizes removed first. That removal is best effort, and everyone with access to the workspace can read what is sent. Agree as a team which repos send content before anyone turns it on.

How it works

Tracing is off everywhere until a developer turns it on for a folder. There are two ways to turn it on:

• /rius:enable-here sends structure only: the span tree, models, tokens, cost, timing, tool names and error types.
• /rius:enable-content-here sends that structure and the content as well.

With content on, each session trace also carries:

• Your prompts and Claude’s replies. The model’s thinking is never sent.
• The session’s name as the trace name, either the name you gave it with /rename or --name, or the title Claude Code generates from your first prompt.
• Tool inputs, such as commands and file paths.
• Tool outputs, such as file contents and command output.
• The error details of a failed tool call.

The decision is made on your machine, before anything is sent. Only you can run the command. Claude can’t run it for you, and a repository’s settings can’t turn tracing on. The rules live in ~/.claude/rius/config.json. They apply to the folder and everything under it, and a disabled folder always wins at any depth. You can enable ~/code with content and still keep one folder inside it out with /rius:disable-here.

Before export, the plugin replaces the secrets it recognizes with a marker such as [redacted:aws-key]. It covers cloud, GitHub, Slack, Stripe, OpenAI, Anthropic and Rius keys, JWTs, private keys, password= style values and Authorization headers. The output of a read of .env*, key and credential files is replaced whole. A secret in a format the plugin doesn’t know is sent. Each captured value is capped at 32 KB and marked where it was cut.

Set it up

1. Install the plugin and sign in. The full walkthrough is in the Claude Code plugin guide.

   claude plugin marketplace add glassflow/rius-coding-agents && claude plugin install rius@rius-coding-agents

   Then run /rius:login, pick the workspace and click Connect.

2. Change into the repo whose prompts and replies you want to record, and start Claude Code from inside it.

3. Enable the folder with content:

   /rius:enable-content-here

   If the session is already open, it is traced from its start, so everything said earlier in it is sent too. To trace only from now on, start a new session after enabling.

4. Check the state:

   /rius:status

   The first line reads Rius tracing: on. The Content: line reads prompts, replies, file contents and command output. A folder enabled with /rius:enable-here reads none (structure only) there instead.

5. Send a prompt, then open the session in the trace list. The prompts, replies and tool output appear on the spans.

6. Settle the team rule before rollout. Each developer enables their own folders, so the agreement has to be a team decision, for example content on in internal tooling repos and structure only in anything holding client data. The team usage guide shows how to share one workspace.

7. To stop content later, run /rius:enable-here in the folder, which drops it back to structure only. To stop it everywhere, set RIUS_CAPTURE_CONTENT to false in the env block of ~/.claude/settings.json and restart Claude Code. The variable can only lower capture, and true is ignored.

   {
     "env": {
       "RIUS_CAPTURE_CONTENT": "false"
     }
   }

What you get

• The full text of each prompt and reply on the spans of the session trace, next to the model calls, tokens and cost.
• Tool inputs and outputs and the error details of failed tool calls, so you can go from a failed session to the exact step and see what the command did.
• A trace named after the session instead of the generic claude-code session.
• Content only from the folders you chose. Structure-only folders in the same workspace still show models, tokens, cost and timing.
• A clear boundary on what leaves your machine. Turning capture off applies from then on, and traces already in the workspace keep their content.

Because redaction is best effort and everyone with workspace access can read the content, treat a content folder as visible to the whole workspace. The Coding agents overview compares how each agent handles content.

When something else fits

• If you only need to see which tools ran, how many tokens a session used and what it cost, /rius:enable-here gives you that with no prompts, replies, file contents or command output sent.
• If a repo holds data that must never leave the machine, run /rius:disable-here in it, or set RIUS_CAPTURE_CONTENT to false to send structure only from every folder.

Related