MiraChat Class

[Bullish-Design]

Organize queries and responses into "flows". Allow results to trigger various actions, and store those results somewhere for context - either user reference or to send to llm.

i.e. - error tracebacks and messages:

• pretty print
• store error and tracebacks
• treesitter search trail using the relevant traceback code for more context
• pull source code or docs from github
• github search for similar code in public repos
• web search
• pull web search source text
• summarize via llm
• "manager update" via llm - track when the error changes and compare against code diffs to track what gets tried. Have llm do a short writeup based on the diff, error, and timing. Also store reference to what the fix was and where I found it.

Etc, etc. Have that run automatically in the background on each test failure, for example. Could then selectively choose what to send to llm with the request to fix, or have presets: "traceback" /"detailed trace" /"trace + github search results" /etc.

Lots of options.