Effort Logging

How It Works

How It Works

Your project grade is based on hours of engaged work, measured from your Claude Code activity. Your project repository configures Claude Code to report each prompt, tool call, and reply to cs124.org the moment it happens, and our server timestamps each one as it arrives. Your hours are computed from that record by a fixed rule, described below. No AI model decides them. Today’s hours update on the /MP page as you work, and each day’s total is saved at the end of the day.

There is nothing to push for your time to count. You do still need to push to save your code—see below.

Getting Connected

Getting Connected

1. Use the Code tab of the Claude desktop app (or claude in a terminal), not the Chat tab and not claude.ai in your browser.
2. Start the session in your project folder: the one containing CLAUDE.md, not the folder you cloned it into. In the Code tab, click + New session, choose Local, then Select folder, and pick your repository’s folder. See the Git guide.
3. Answer yes if asked about trust. If Claude asks whether you trust the files in this folder, answer yes. That is what enables the project’s hooks, and without it nothing is reported.
4. Send a prompt. Nothing is reported until you do.

Once the first event arrives, the /MP page shows “Claude Code activity received”. If it still says Claude Code isn’t reporting, check that you started the session in the right folder, and that .claude/settings.json in your repository is unmodified. Asking Claude to run git checkout .claude/settings.json, or running it yourself in the terminal pane, restores it.

How Hours Are Counted

How Hours Are Counted

Your hours are made of turns. A turn starts when you send a prompt and ends when you send the next one. Each turn earns:

• The time Claude spends working on it, up to ten minutes.
• The time it takes to read the reply, at 1,200 characters a minute, counting both Claude’s reply and the code changes it shows you.
• The time it takes to type your next prompt, at 200 characters a minute.
• Two minutes to think.
• Up to two minutes for each permission dialog Claude is waiting on you to answer.

A turn never earns more than the time that actually passed before your next prompt. Your daily hours are the sum of your turns.

Here is one turn. You ask Claude to add a button to your main screen. Claude works for three minutes, edits a file, and replies; the reply and the change it shows you come to 1,200 characters. You read it, run the app, and nine minutes after your first prompt you send a 400-character follow-up describing what you saw. That turn earns three minutes of Claude working, one minute of reading, two minutes of typing, and two minutes to think: eight minutes of the nine that passed.

Some time does not count:

• Walking away during a long task. Claude’s working time counts for at most ten minutes per prompt, whether or not you watch the rest.
• Leaving a permission dialog open. Each dialog earns at most two minutes, and Claude is not working while it waits for you.
• Time away from the conversation. Once the allowances above are used up, time before your next prompt earns nothing.
• Testing your app or reading on your own beyond the thinking allowance. Type what you saw into your next prompt; that time counts.

What Counts

What Counts

Only work on your CS 124 project counts. That means Claude Code sessions started in your project folder, spent on your app: building it, fixing it, testing it, deciding what to build next, and learning what you need to do those things.

These do not count, and do not belong in your project folder:

• Work for other courses, including CS 124 homework.
• Other projects of your own, even ones built with Claude Code.
• Questions unrelated to your app, however useful.

Use a different folder, or a Chat conversation, for anything that is not your project. Staff review session transcripts, and time spent on anything else is removed from your hours.

What Is Sent and When

What Is Sent and When

• On each prompt: the text of your prompt.
• Before each tool call: the tool name and a short summary of its input.
• After each tool call: the tool name and a short, truncated summary of its input and result—file paths, commands, and edited content.
• When Claude asks permission to use a tool, and when a request is refused: the tool name and a short summary of what it wants to do.
• When Claude Code shows a notification: its type and message, such as a permission dialog waiting for you or Claude waiting for your next prompt.
• When Claude finishes a reply, or stops on an error: the final reply text, or the kind of error.
• When a subagent starts or stops, and when the conversation is compacted: a marker.
• When a session starts or ends: a marker.

Each report also carries the session id and the path of your project folder. Everything is sent over HTTPS to cs124.org, is visible to course staff only, and is the same material as the transcript already saved in your repository.

Still Commit and Push

Still Commit and Push

Pushing is how your code is saved, and it also keeps a backup copy of each session transcript in .claude-sessions/ in your repository. If reporting ever fails—no network, or something blocking the hooks—the pushed transcript is used instead, so nothing is lost as long as you push. Commit and push at the end of every work session.

Use Claude Code for Everything

Use Claude Code for Everything

This includes work that happens outside your code editor. If your app uses external services (like Firebase, Supabase, a REST API, etc.), ask Claude Code for step-by-step instructions rather than following a tutorial on your own. For example:

• Instead of reading the Supabase docs yourself, ask Claude: “I need to set up a Supabase project for my app. Walk me through the steps.”
• Instead of configuring Firebase in the console directly, ask Claude: “Help me set up Firebase Authentication for my app. What do I need to do in the Firebase console and in my code?”
• Instead of searching Stack Overflow, ask Claude: “My list of saved items stops scrolling after the first screen. What’s wrong?”

This way the conversation is reported—even if the actual work happens in a browser or dashboard—and we can see the time you spent figuring things out.

Common Mistakes

Common Mistakes

• Using the Chat tab or claude.ai instead of Claude Code: Neither reports anything. Use the Code tab of the Claude desktop app, or claude in a terminal.
• Working in the wrong folder: The hooks belong to your project, so a session started anywhere else is not reported. That includes the CS 124 folder you cloned into, which is the parent of your project. Start a new session in the folder containing CLAUDE.md.
• Declining trust: Answering no when asked whether you trust the files in this folder disables the hooks. Start a new session in your project folder and answer yes.
• Editing files in .claude/: Changing .claude/settings.json can break or misdirect reporting. Leave it alone, and run git checkout .claude/settings.json if it has changed.
• Working without Claude: If you spend an hour debugging by reading documentation on your own, that time is invisible. Route your questions through Claude Code instead.
• Long idle sessions: Having Claude Code open while you’re not actively working doesn’t count as effort. We measure engaged interaction, not idle time.

Checking Your Activity

Checking Your Activity

The /MP page shows when your last Claude Code activity was received, how many sessions you’ve had today, your hours today so far, your hours for each week, and your progress toward the 24 hours the project requires. If the last activity received is older than the session you just finished, your hooks aren’t reporting—go back to getting connected. The sessions page lists every session by day, with each turn of the conversation and the minutes it earned.

Working on your project with Claude Code during your Tuesday discussion section counts toward your hours, measured the same way as any other time. Attendance by itself earns no hours.