Hour a Day: AI - Day 2 - Starting off with a plan

In yesterday’s post, I got all the blog’s dependencies updated and the blog building with the help of my new best friend Claude + Aider.

It did take a substantial amount of time to format the blog though with all the prompts involved and manually copying and pasting from the history file.

So I think the first thing to do as part of this adventure is make writing about the adventure a lot easier! Let’s automate it!

The Initial Requirements doc

I have heard a lot that when working with generative AI it’s best to have clear requirements defined in a file that can be used as a general prompt throughout the building process. Why lose all the valuable information when the chat session ends?

So I started with this small requirements doc, not trying too hard to get details finalized yet.

High Level Goal for “llm-history-to-post”
llm-history-to-post is an application that will parse llm chat logs and create blog post templates based on user input.
Requirements
The application should interact over the cli
The application should parse a “.aider.chat.history.md” into an in memory collection of the user entered prompt and the llm output response
This parsed input should be logically organized by day
When the application is run the user should be able to select which singular day they would like to process
After the user has selected the desired day, the user should be presented a list of the parsed prompts where they can choose to select which prompts / responses to populate into the end result
The user should be able to select any number of prompts to process while skipping others.
This should be a multi select list offered like SharPrompt
The user should select the prompt response pair by only previewing the prompt itself.
Once selected, each prompt response pair should be represented to the user and then asked if verdict was pass or failure. This input should make it into the blog.
Once all the questions have been answered the application should output a new blog post that was constructed from a source template

So I switched over to aider’s ask mode and asked it for feedback on these requirements

Prompt:
can you look at requirements.md and help me fine tune the individual asks to better suit feeding into aider for implementation?

Response:
llm-history-to-post/requirements.md
Add file to the chat? (Y)es/(N)o/(D)on’t ask again [Yes]: y
After reviewing the requirements.md file, I can help refine the requirements to make them more suitable for implementation with aider. Here’s my analysis and suggestions:
Refined Requirements for llm-history-to-post
CLI Interface
This is clear, but we could specify the command structure, e.g., llm-history-to-post [options] [path-to-history-file]
Consider adding optional flags like --output-dir or --template-path
Parsing Chat History
The format of “.aider.chat.history.md” should be more explicitly defined
Specify how to identify user prompts vs LLM responses in the file
Define what “logically organized by day” means - are there date headers in the file?
Day Selection
Specify the UI for day selection (numbered list?)
What if there’s only one day in the history?
What format should dates be displayed in?
Prompt Selection
The multi-select with SharpPrompt is clear
Specify how much of each prompt to preview (first N characters?)
Should there be an option to view the full prompt before selecting?
Pass/Failure Verdict
Clarify the format for collecting this information (Y/N prompt? Multiple choice?)
Define what “pass” and “failure” mean in this context
Specify how this verdict should be formatted in the blog post
Blog Post Generation
The “source template” is mentioned but not defined - specify the template format
Define the output format (Markdown? HTML?)
Specify where the generated blog post should be saved
Consider adding metadata like title, date, tags
Error Handling
Add requirements for handling invalid input files
Specify behavior when user cancels mid-process
Define how to handle empty selections
Additional Considerations
Add ability to edit prompts/responses before inclusion in the blog
Consider pagination for large history files
Add option to save selection state for resuming later
By making these requirements more specific and detailed, it will be easier for aider to generate accurate implementation code that meets your needs precisely.
Would you like me to rewrite the entire requirements document with these refinements, or would you prefer to focus on specific areas?
Tokens: 1.0k sent, 497 received. Cost: $0.01 message, $0.01 session.
^C again to exit

Verdict: ✅ Exactly what I was looking for! It had questions that I knew I hadn’t specified but I wanted to see if it understood the general idea before getting more technical.

Taking Suggestions!

Prompt:
yes please help refine the requirements. you can look at 2025-03-27-hadai-day-1.md as the source to answer some of your questions about formatting.

Response:
content/post/2025/03/2025-03-27-hadai-day-1.md
Add file to the chat? (Y)es/(N)o/(D)on’t ask again [Yes]: y
Refined Requirements for “llm-history-to-post”
High Level Goal
llm-history-to-post is an application that parses LLM chat logs and creates blog post templates based on user input, following the format seen in the example blog post.
Detailed Requirements
CLI Interface
The application should be invoked via command line: llm-history-to-post [path-to-history-file]
Optional flags:
--output-dir: Directory to save the generated blog post (default: current directory)
--template-path: Path to a custom blog post template (default: built-in template)
Parsing Chat History
Parse a “.aider.chat.history.md” file into an in-memory collection of user prompts and LLM responses
The file format contains sections with:
User prompts prefixed with “> Prompt:”
LLM responses prefixed with “> Response:”
Group these prompt-response pairs by day based on timestamps in the file
Each day should be identified by a date in the format YYYY-MM-DD
Day Selection
Present the user with a numbered list of available days found in the history file
Display dates in YYYY-MM-DD format
If only one day exists, automatically select it
Allow user to select a single day to process
Prompt Selection
After day selection, display a multi-select list (using SharpPrompt or similar) of all prompts from that day
Show the first 100 characters of each prompt as a preview, with “…” if truncated
Allow selecting any number of prompt-response pairs for inclusion in the blog post
Provide an option to view the full prompt before selecting (optional)
Verdict Collection
For each selected prompt-response pair:
Display the full prompt and response to the user
Ask if the verdict was “pass” or “failure” with a simple prompt: “Was this a success? (Y/N)”
Store the verdict with the prompt-response pair
Format verdicts in the blog post as:
Success: “Verdict: ✅ [user comment]”
Failure: “Verdict: ❌ [user comment]”
Blog Post Generation
Use a template similar to the example blog post (2025-03-27-hadai-day-1.md)
Include:
YAML frontmatter with title, date, categories, and tags
Introduction section (can be static or prompt user for input)
Selected prompt-response pairs formatted as in the example:
Prompt in a blockquote with “> Prompt:” prefix
Response in a blockquote with “> Response:” prefix
Verdict with emoji (✅/❌) and user comment
Conclusion section (can be static or prompt user for input)
Generate the post as a Markdown file with naming convention: YYYY-MM-DD-hadai-day-N.md
Save to the specified output directory or current directory
Error Handling
Gracefully handle invalid or missing history files
Allow cancellation at any point in the selection process
Handle empty selections (warn user and confirm before proceeding)
Validate that the output directory exists and is writable
Additional Features
Allow user to edit the prompt/response text before inclusion in the blog
Provide a summary of selections before generating the final blog post
Option to save selection state for resuming later
Support for custom templates with variable substitution
This refined requirements document provides clearer specifications based on the example blog post format and addresses the questions raised in my previous analysis.
Tokens: 4.9k sent, 744 received. Cost: $0.03 message, $0.04 session.

Verdict: ✅ While it didn’t give me the EXACT requirements I wanted, and it incorrectly called out the format of the history file, it did give me a great template to modify to specify the actual requirements I wanted.

Requirements V3

After some manual editing here’s what I landed on for today.

LLM History to Post
High Level Goal
llm-history-to-post is an application that parses LLM chat logs and creates blog post templates based on user input, following the format seen in the example blog post.
Detailed Requirements
CLI Interface
The application should be invoked via command line: llm-history-to-post [path-to-history-file]
There are no flags specified
The input file is stored in “.aider.chat.history.md” which resides in the current directory
The output file will be specified as the path content/posts/{year}/{month}/{year}-{month}-{day}-hadai-day-.md
Parsing Chat History
Parse a “.aider.chat.history.md” file into an in-memory collection of user prompts and LLM responses
The file format contains sections with:
User prompts prefixed with “####”
LLM responses are the gaps in-between prompts
Group these prompt-response pairs by day based on timestamps in the file. Each line won’t have a timestamp but there will be section headings like “# aider chat started at 2025-03-26 19:18:06”
Each day should be identified by a date in the format YYYY-MM-DD
Day Selection
Present the user with a numbered list of available days found in the history file
Display dates in YYYY-MM-DD format
If only one day exists, automatically select it
Allow user to select a single day to process
Prompt Selection
After day selection, display a multi-select list (using SharpPrompt or similar) of all prompts from that day
Show the first 100 characters of each prompt as a preview, with “…” if truncated
Allow selecting any number of prompt-response pairs for inclusion in the blog post
Provide an option to view the full prompt before selecting (optional)
Verdict Collection
For each selected prompt-response pair:
Display the full prompt and response to the user
Ask if the verdict was “pass” or “failure” with a simple prompt: “Was this a success? (Y/N)”
Store the verdict with the prompt-response pair
Format verdicts in the blog post as:
Success: “Verdict: ✅ [user comment]”
Failure: “Verdict: ❌ [user comment]”
Blog Post Generation
Use a template similar to the example blog post (2025-03-27-hadai-day-1.md)
Include:
YAML frontmatter with title, date, categories, and tags
Introduction section (can be static or prompt user for input)
Selected prompt-response pairs formatted as in the example:
Prompt in a blockquote with “> Prompt:” prefix
Response in a blockquote with “> Response:” prefix
Verdict with emoji (✅/❌) and user comment
Conclusion section (can be static or prompt user for input)
Generate the post as a Markdown file with naming convention: YYYY-MM-DD-hadai-day-N.md
Save to the specified output directory or current directory
Error Handling
Gracefully handle invalid or missing history files
Allow cancellation at any point in the selection process
Handle empty selections (warn user and confirm before proceeding)
Validate that the output directory exists and is writable
Additional Features
Support for custom templates with variable substitution

This was pretty good progress for today! A lot of my “hour” is spent getting these posts formatted appropriately so I’m looking forward to getting this automated. Let’s see how that plays out tomorrow!