└── README.md


/README.md:
--------------------------------------------------------------------------------
 1 | # llm-prompts
 2 | 
 3 | A collection of prompts for use with the LLM CLI tool
 4 | 
 5 | Relevant:
 6 | 
 7 | - [LLM documentation](https://llm.datasette.io/)
 8 | - [llm, ttok and strip-tags—CLI tools for working with ChatGPT and other LLMs](https://simonwillison.net/2023/May/18/cli-tools-for-llms/)
 9 | - [symbex: search Python code for functions and classes, then pipe them into a LLM](https://simonwillison.net/2023/Jun/18/symbex/)
10 | 
11 | Many of these are "system prompts" - where content is piped to the tool along with a `--system prompt` that tells the tool how it should process that data.
12 | 
13 | ## Using symbex
14 | 
15 | [symbex](https://github.com/simonw/symbex) is a CLI tool for searching for and outputting subsets of Python code based on search criteria. The tool is designed to help extract content for passing to `llm`.
16 | 
17 | ## Features derived from tests
18 | 
19 | > Detailed bullet point list of features of this software
20 | 
21 | Running this in a fresh checkout of [simonw/symbex](https://github.com/simonw/symbex) with `3.5-16k` - the 3.5 turbo model with a 16,000 token context:
22 | 
23 | ```bash
24 | symbex 'test_*' | llm --system 'Detailed bullet point list of features of this software' -m 3.5-16k
25 | ```
26 | Output:
27 | 
28 | > Features of this software include:
29 | > 
30 | > 1. Testing of command-line interface to ensure it shows help when no arguments are provided.
31 | > 2. Testing of command-line interface to ensure it returns the expected output for different argument combinations.
32 | > 3. Testing of command-line interface to ensure it shows errors when necessary.
33 | > 4. Testing of reading a file with a specific encoding to ensure correct text is read.
34 | > 5. Testing of formatting quoted strings to ensure they are properly escaped and formatted.
35 | > 6. Testing of symbol extraction to ensure correct symbols are identified from the given code.
36 | > 7. Testing of method symbols to ensure correct methods are identified from the given code.
37 | > 8. Testing of retrieving symbols with documented and undocumented filters to ensure correct symbols are returned.
38 | > 9. Testing of filtering symbols based on various options such as function, class, async, typed, untyped, partially-typed, fully-typed, documented, and undocumented.
39 | > 10. Testing of counting the number of symbols returned by the filters.
40 | 
41 | The "Testing of" bits there indicate it didn't fully understand the idea of describing the software, not the tests.
42 | 
43 | GPT-4 handled this much better:
44 | 
45 | ```bash
46 | symbex 'test_*' | llm --system 'Detailed bullet point list of features of this software' -m 4
47 | ```
48 | > - Search for symbols (functions, classes, etc.) in a given codebase
49 | > - Show detailed information on the selected symbol (e.g. function signature)
50 | > - Filter symbols based on various criteria such as:
51 | >   - Function
52 | >   - Class
53 | >   - Async
54 | >   - Typed
55 | >   - Partially typed
56 | >   - Fully typed
57 | >   - Documented
58 | >   - Undocumented
59 | > - Support file and directory filtering
60 | > - Handle different Python versions and file encodings
61 | > - Include tests for functionality and CLI behavior
62 | > - Quoted string support for single and multiline strings
63 | > - Count the number of matching symbols
64 | > - CLI Runner to execute command line interface tests
65 | > - Parametrized testing for multiple input-output combinations
66 | > - Display line numbers for symbol, classes, methods, etc.
67 | 
68 | 


--------------------------------------------------------------------------------