Dark Mode
The Sonar Chronicles
in the style of Dan Brown
Chapter 3 of 8
35%
Chapters
I · The Blueprint
scaffold project · Mar 15
II · The Extraction
add extractor · Mar 16
III · The Mind
add analyzer · Mar 17
IV · The Voice
add narrator · Mar 18
V · The Wiring
wire CLI · Mar 19
Previously: The extractor learned to parse raw commit data — counting files, detecting languages, and pulling function signatures from diffs. The system could see, but it couldn't yet understand.
Chapter III
The Mind

The diff was sixty-three lines long, but its true weight couldn't be measured in lines of code. Dr. Sarah Chen stared at the terminal, knowing that what she was about to build would give the system something no parser could provide — intent.

async function analyze(diff: string, facts: Facts): Promise<IR> { const response = await claude.messages.create({ model: "claude-sonnet-4-20250514", messages: [{ role: "user", content: buildPrompt(diff, facts) }] }); return parseIR(response); }

The function was deceptively simple. But inside that buildPrompt call lay the real architecture — a carefully constructed prompt that would teach the LLM to see what humans see when they read code.

interface IR { intent: string; // What the developer was trying to achieve architecture: string[]; // Structural decisions revealed decisions: Decision[]; // Why this approach over alternatives narrative_hooks: string[]; // What makes this commit interesting }

Sarah ran the analyzer against the test commit. The terminal filled with structured JSON — not a story yet, but the skeleton of one. The mind could think. Now it needed a voice.

Previous← II · The Extraction
Light Mode
The Sonar Chronicles
in the style of Dan Brown
Chapter 3 of 8
35%
Chapters
I · The Blueprint
scaffold project · Mar 15
II · The Extraction
add extractor · Mar 16
III · The Mind
add analyzer · Mar 17
IV · The Voice
add narrator · Mar 18
V · The Wiring
wire CLI · Mar 19
Previously: The extractor learned to parse raw commit data — counting files, detecting languages, and pulling function signatures from diffs. The system could see, but it couldn't yet understand.
Chapter III
The Mind

The diff was sixty-three lines long, but its true weight couldn't be measured in lines of code. Dr. Sarah Chen stared at the terminal, knowing that what she was about to build would give the system something no parser could provide — intent.

async function analyze(diff: string, facts: Facts): Promise<IR> { const response = await claude.messages.create({ model: "claude-sonnet-4-20250514", messages: [{ role: "user", content: buildPrompt(diff, facts) }] }); return parseIR(response); }

The function was deceptively simple. But inside that buildPrompt call lay the real architecture — a carefully constructed prompt that would teach the LLM to see what humans see when they read code.

interface IR { intent: string; // What the developer was trying to achieve architecture: string[]; // Structural decisions revealed decisions: Decision[]; // Why this approach over alternatives narrative_hooks: string[]; // What makes this commit interesting }

Sarah ran the analyzer against the test commit. The terminal filled with structured JSON — not a story yet, but the skeleton of one. The mind could think. Now it needed a voice.

Previous← II · The Extraction