Vijay Janapa Reddi
dcac6d8979
Removed orchestration code, prompt templates, and one-off LLM scripts
that were either zero-referenced, self-referencing, or already retired:
book/tools/audit/
- subagent_lane.py + subagent_prompts/sentence_case_h3.md
(audit-driver scaffolding, no external callers)
book/tools/scripts/
- gemini_review.py, figure_audit_gemini.py
- testing/gemini_math_check.py
(zero external references; one-off audit utilities)
scripts/cross-references/
- agent-playbook.md (prompt template; only referenced by one
print line in audit_crossrefs.py, scrubbed)
periodic-table/scripts/archive/
- entire directory (14 files): iteration-loop scripts and
panel-debate runners used to produce v0.2 of the periodic
table. README labeled it research provenance only; not part
of any active build.
No CI workflows, Makefiles, or active scripts touch any of the
removed files. Net: -2,523 lines.
Figure Audit Automation
This directory contains figure_audit.py, a script designed to automate the visual auditing of figures within the ML Systems textbook.
What it does
The script orchestrates a multimodal audit of every figure across Volume 1 and Volume 2 of the textbook. It ensures that the prose, the captions (fig-cap), and the alt-text (fig-alt) precisely match the content of the fully rendered visual images.
- Discovery: It scans the
book/quarto/contents/directory to identify all.qmdchapters containing figures. - Visual Extraction: It resolves the corresponding published HTML URL for each chapter, parses the HTML, and downloads the exact rendered
<img src="...">and inline<svg>visual assets locally. - Auditing: It dispatches parallel worker tasks via the
geminiCLI. The CLI is given explicit instructions to load the local images visually, compare them directly against the.qmdsource text, and evaluate them based on thefigure-audit-brief.mdrubric. - Reporting: It generates strict, granular YAML output files in
.claude/_reviews/Figure Audit/, detailing any misalignments (e.g., the text claims10^4but the chart shows10^3) along with surgically precise.qmdfix recommendations.
How to use it
Run the script from the repository root:
python3 scripts/figure_audit.py
Pre-requisites
- You must have
geminiCLI installed and authenticated on your local machine. - The script assumes the rendered HTML book is available at
https://harvard-edge.github.io/cs249r_book_dev/...(used purely to scrape the final image variants).
Applying the fixes
Once figure_audit.py finishes running, your .claude/_reviews/Figure Audit/ directory will be populated with .yml files containing proposed_fix entries.
These fixes are written as precise, minimal adjustments targeting the .qmd source files. They can either be applied manually by a human reviewing the YAML reports, or parsed programmatically/agentically to apply the diffs across the workspace.