Claude Code: Keyboard Shortcuts and Daily Usage Patterns

Who this is for: Researchers who have installed Claude Code and want to use it fluently — not just knowing what it can do, but developing the habits that make it fast and comfortable day to day.

The essential shortcuts

These work inside a running Claude Code session:

The most important: Ctrl+C. If Claude is doing something wrong — heading in the wrong direction, taking too long, misunderstanding the task — interrupt immediately. Then rephrase and try again. Do not wait for it to finish.

Slash commands

Commands starting with / are meta-operations — they control Claude Code itself rather than asking Claude to do something.

/clear vs Ctrl+L: /clear wipes the conversation history — Claude forgets everything from this session. Ctrl+L just clears the visual display; the conversation is still there. Use /clear when you are starting a genuinely new task and do not want previous context interfering.

/compact: When a long session is approaching context limits, /compact asks Claude to summarise the conversation so far and replace the full history with the summary. This frees up space for more work without losing the thread entirely. Useful for long batch jobs.

/rewind: Every action Claude takes creates a checkpoint. If a batch operation goes wrong — Claude processed files in the wrong format, wrote to the wrong location, made a mistake partway through a multi-step task — /rewind lets you restore the conversation and project state to a previous checkpoint. Think of it as undo for Claude Code sessions. Useful when you want to try a different approach from a known good state, rather than correcting a sequence of wrong actions.

Launching Claude Code well

The habit that matters most: start from the right folder.

cd /path/to/your/project
claude

This puts Claude in the context of your project. It will find your CLAUDE.md automatically, and you can refer to files by short names rather than full paths.

Shortcut on Windows: Navigate to the folder in File Explorer, then type powershell in the address bar — it opens a terminal already in that folder. Then type claude.

Shortcut on Mac: Drag a folder onto the Terminal icon, or right-click a folder in Finder → "New Terminal at Folder" (may require enabling in System Preferences → Keyboard → Services).

Starting a session with context

Instead of just > what do you need?, begin sessions with orientation:

> We are continuing work on the Montaillou deposition corpus.
  Read CLAUDE.md first, then tell me where we left off according to progress.md.

Or, if you have a CLAUDE.md with a progress log, simply:

> Read CLAUDE.md and progress.md and let's continue.

This takes 10 seconds and means Claude immediately understands the project, the conventions, and the current state. The alternative — explaining context from scratch each session — is slower and results in worse calibration.

Writing good prompts: the habits that matter

Be specific about format:

> Summarise paper.pdf as: main argument (2 sentences), key evidence (bullet list),
  methods (1 sentence), critical note (1 sentence).

vs.

> Summarise paper.pdf

The first gets you a usable structured note. The second gets you something you will need to reshape.

Specify where to save output:

> Read paper.pdf and write the summary to notes/paper_summary.md

If you do not specify, Claude outputs to the terminal — which you then have to copy-paste. Specifying a file path means the output lands where you need it immediately.

Give constraints when you have them:

> Max 200 words. Chicago author-date citations. Academic register.

Stated upfront, these shape the entire response. Added as corrections after the fact, they require regeneration.

If the task is complex, describe the end state:

> I want to end up with a CSV file called extractions.csv where each row
  is one deposition, with columns: deponent, date, location, inquisitor.
  The source files are the DOCX files in /transcriptions/.

Claude will figure out the steps. You describe the destination; Claude plans the route.

Multi-step tasks: letting Claude work

For tasks that involve several operations — read multiple files, process, consolidate, write output — you can hand off the whole thing:

> There are 12 PDFs in /sources/.
  For each one:
  1. Read it
  2. Extract all named persons with page references
  3. Note the document's date range if given
  Write everything to extractions.md, one section per file.

Claude will work through this sequentially. You can watch it in the terminal — the ● indicators show each tool call. You do not need to supervise every step.

When to stay in the loop: If you are unsure whether Claude is interpreting the task correctly, add: "After reading the first file, show me the output before continuing, so I can confirm the format is right." This gives you a checkpoint before Claude processes all 12.

Iterating in a session

Claude Code sessions work well for iterative refinement:

> Read draft.docx and give me feedback on the introduction.

[Claude responds]

> The third point about methodology — expand on that. Also, the opening
  sentence is too weak. Suggest 3 alternatives.

[Claude responds]

> Use option 2 for the opening. Now write the revised introduction to
  draft_revised.docx

The conversation history is the working space. Claude remembers what was said earlier in the session — you do not re-explain, you iterate. This is the "thinking partner" mode that researchers find most valuable once they develop the habit.

Handling interruptions in your day

Pausing a task mid-session: If you need to stop and come back — a meeting, end of day — you cannot "save" a Claude Code session. But you can:

> Before I close this session: summarise what we have done and what
  remains to do. Save to notes/session_log.md

Then when you return, start with:

> Read notes/session_log.md and let's continue.

Starting fresh vs. continuing: Use /clear when the current task is done and you are starting something unrelated. Keep the same session for related work — the accumulated context is useful and means less re-explanation.

Useful daily habits

Keep a progress.md in each project. Ask Claude to update it at the end of a work session:

> Append a brief note to progress.md summarising what we did today
  and what the next step is.

Ask Claude to explain before it acts. For any task you are uncertain about:

> Before you start, describe what you plan to do, step by step.
  Wait for my confirmation before proceeding.

Use /cost to stay aware. At the end of a long session, /cost shows how many tokens were used. Useful for understanding which types of tasks are token-heavy and calibrating your work accordingly.

Name your output files descriptively. Instead of output.md, use montauban_1241_firstpass.md. Claude will use the name you give it — make it something useful to you when you find the file two weeks later.

When Claude gets confused or goes wrong

Interrupt immediately (Ctrl+C) — do not wait for it to finish and then correct.

Rephrase, do not just repeat. If Claude misunderstood, the same prompt will produce the same misunderstanding. Change the framing:

> I didn't mean summarise the whole document — I only want the section
  on methodology, pages 12–18.

Use /clear and start over if the conversation has accumulated too much wrong context. Sometimes a fresh start with a better initial prompt is faster than trying to correct course over many exchanges.

Be explicit about what was wrong:

> The names list you produced included names from the inquisitor's
  questions, not only from the deponent's answers. Redo the extraction,
  only taking names mentioned by the deponent, not by the inquisitor.

Specific corrections produce specific improvements. "This is not what I wanted" does not.

Official documentation: code.claude.com/docs — full command reference, shortcuts, and advanced usage patterns.

Related

• A3.code-basics-non-programmers — installation and first launch

• A9.markdown-project-memory — CLAUDE.md and progress.md setup

• A.setup.settings-local — permissions configuration

• A4.conceptual-vocabulary — understanding what /compact and context limits mean