Changelog - 2026-03-23_01

Assure Proper Integration of Table of Contents

The Table of Contents was consistently missing from final memo output despite the toc_generator agent running successfully mid-pipeline. Two separate ordering bugs caused the TOC to either never generate or get destroyed before finalization.

Why

Investment memos require a navigable Table of Contents for both HTML/PDF exports and Obsidian viewing. Every run since the TOC agent was introduced has produced a final draft without a TOC — despite the agent logging "TOC generated" in the console. Users had to manually run cli/assemble_draft post-pipeline to get a TOC.

What Changed

Bug 1: TOC Agent Ran Before Final Draft Existed

The workflow graph placed the toc node BEFORE assemble_citations, which is the agent that actually creates the final draft file. The TOC agent looked for a final draft, found nothing, and skipped.

Before:

... → enrich_visualizations → toc → revise_summaries → ... → assemble_citations → ...
                               ↑                                      ↑
                          No file exists yet              Creates the file here

After:

... → enrich_visualizations → revise_summaries → ... → assemble_citations → toc → ...
                                                              ↑                ↑
                                                     Creates the file    Reads and inserts TOC

Commit: d87fa36

Bug 2: integrate_scorecard Destroyed the TOC During Reassembly

Even after fixing Bug 1, the TOC was still missing. integrate_scorecard (the last agent that rewrites the final draft) attempted to import cli.assemble_draft.assemble_final_draft — which includes TOC generation. But cli/ had no __init__.py, making it unimportable. The import silently failed, falling through to a basic concatenation fallback with no TOC.

# This ALWAYS raised ImportError because cli/ wasn't a package
try:
    from cli.assemble_draft import assemble_final_draft  # includes TOC
    ...
except ImportError:
    # Fallback: raw concatenation, NO TOC
    for section_file in sorted(sections_dir.glob("*.md")):
        content += section_file.read_text() + '\n\n'

Fix: Created cli/__init__.py and rewrote the fallback to call assemble_citations() + toc_generator_agent().

Commit: cbd19c8

TOC Agent Made Context-Aware

Previously, the TOC agent blindly skipped if any ## Table of Contents heading existed. Now it validates accuracy and location:

1. Strips existing TOC from content

2. Generates correct TOC from current headers

3. Compares: if existing TOC matches and is correctly positioned → no changes

4. Otherwise: regenerates and reinserts in the correct location

Commit: d87fa36

Verification

Files Changed