Add Stash The Current Prompt To Send Another First as a Claude Code TIL

README.md

@@ -10,7 +10,7 @@ working across different projects via [VisualMode](https://www.visualmode.dev/).
 
 For a steady stream of TILs, [sign up for my newsletter](https://visualmode.kit.com/newsletter).
 
-_1772 TILs and counting..._
+_1774 TILs and counting..._
 
 See some of the other learning resources I work on:
 
@@ -168,6 +168,7 @@ If you've learned something here, support my efforts writing daily TILs by
 - [Monitor Usage Limits From CLI](claude-code/monitor-usage-limits-from-cli.md)
 - [Open Current Prompt In Default Editor](claude-code/open-current-prompt-in-default-editor.md)
 - [Resume Specific Session](claude-code/resume-specific-session.md)
+- [Stash The Current Prompt To Send Another First](claude-code/stash-the-current-prompt-to-send-another-first.md)
 
 ### Clojure
 
@@ -716,6 +717,7 @@ If you've learned something here, support my efforts writing daily TILs by
 
 ### LLM
 
+- [Count Number Of Tokens In A File](llm/count-number-of-tokens-in-a-file.md)
 - [Send cURL To Claude Text Completion API](llm/send-curl-to-claude-text-completion-api.md)
 - [Use The llm CLI With Claude Models](llm/use-the-llm-cli-with-claude-models.md)
 

claude-code/stash-the-current-prompt-to-send-another-first.md

@@ -0,0 +1,25 @@
+# Stash The Current Prompt To Send Another First
+
+I've been working my way through the current cohort of Matt Pocock's [Claude
+Code for Real
+Engineers](https://www.aihero.dev/cohorts/claude-code-for-real-engineers-2026-04).
+The best part about going through a series of videos like this is being able to
+pick up big and small tips and tricks from another person's workflow.
+
+One of the small things I picked up in an early video is the ability to stash
+the current prompt.
+
+Let's say I've gone to the trouble of writing out a detailed prompt, `@`'ing
+some files, and so forth. Then I realize I need first prompt Claude to do
+something else first. Instead of copy-pasting that prompt into my notes,
+deleting it, issuing a different prompt, and then pasting it back in, I can hit
+`Ctrl-s`.
+
+`Ctrl-s` will _stash_ the current prompt, clearing out the prompt input. I can
+then type in something else. Once I hit enter for that new prompt, it will be
+sent to Claude and the stashed prompt will be immediately populated back into
+the input.
+
+Though `Ctrl-s` is mentioned when you hit `?` from within `claude` session, I
+don't see it documented anywhere in their [Interactive Mode
+reference](https://code.claude.com/docs/en/interactive-mode).

llm/count-number-of-tokens-in-a-file.md

@@ -0,0 +1,26 @@
+# Count Number Of Tokens In A File
+
+Over time you have accumulated a bunch of small directives, corrections, and
+project details in your `CLAUDE.md` or `AGENTS.md` file. The file doesn't seem
+too big, but you are mindful that it is being included in every prompt. How many
+tokens is it eating from the context window?
+
+OpenAI's BPE (Byte Pair Encoding) tokenization library,
+[`tiktoken`](https://github.com/openai/tiktoken), is an open-source Python
+package. If it is installed on our machine, then we can use it as part of the
+following one-liner to check a file:
+
+```bash
+❯ python -c "import tiktoken, sys; print(len(tiktoken.encoding_for_model('gpt-4o').encode(open(sys.argv[1], 'r', encoding='utf-8').read())))" \
+    AGENTS.md
+1018
+```
+
+I ran this against the `AGENTS.md` file in a team project I'm on. It came out to
+1018 tokens. This is a very good approximation based on the tokenizer trained
+for `gpt-4o`. The tokenizers may vary a little from model to model, but the
+differences for our purposes here are going to be negligible.
+
+This one-liner gets the "first" argument to the command, reads it in, and runs
+that string against the tokenizer. The length of the tokenized encoding is then
+printed.