[RaghavChamadiya]

What

• One file count. With repo="all", the count for a repo used to differ from the count in that repo's own overview. Each repo row now states code_files from the same population as code_health.code_files in the single-repo overview. That population is every code file health read, scored or not, after the health exclude rules. total_files becomes total_code_files.
• Real summaries. Each repo="all" row now has the overview essay's first sentence (up to 140 characters) instead of its heading, which was "Project Summary" for every repo.
• list_repos stays current. When list_repos re-reads the workspace config, it applies the workspace state sync in memory. It then reports each repo's own state.json values without waiting for a CLI command to rewrite the config.

Why

The count for a repo used to include external package nodes, so one repo showed three different totals depending on where you looked: 6,622, 6,036 and 5,868 on a nine-package monorepo. The summary field never carried a sentence.

How

• A new crud helper, count_health_files, counts the health population without loading the rows.
• Before health has run, a repo row falls back to the graph's code-language files.
• _lead_sentence skips headings, tables, lists, quotes and code fences before it takes the first sentence.

Verification

	Before	After
repo="all" size (chars)	1,384	1,760
Monorepo count: repo="all" / its overview	6,622 / 5,868	5,884 / 5,884
Python service: repo="all" / its overview	1,552 / 698	1,156 / 1,156

• Tests are added for count agreement, lead-sentence extraction and the list_repos state sync.
• pytest tests/unit/server tests/unit/workspace passes (7,601 tests).
• ruff check . is clean.