GenerateTelosSummary: H2 "## ID: title" entries with multi-paragraph bodies silently fragment into fake sequential items

[waveman2020-sudo]

File: LifeOS/install/LIFEOS/TOOLS/GenerateTelosSummary.ts

Environment: LifeOS 7.1.1, Claude Code 2.1.206, Bun 1.3.11, Ubuntu 24.04.4 LTS, ThinkPad (AMD Ryzen AI 9 HX PRO 370, 24 threads, 86GB RAM).

Summary

parseItems() recognizes typed-ID TELOS entries in two forms only: bullets (- **ID**: text) and H3 headings (### ID: text). A source file using H2 headings (## T0: title) with a multi-paragraph prose body underneath — the shape produced by a voice-capture/raw-transcript convention, one heading, several paragraphs — matches neither regex. items.length comes back 0, so the code falls through to the documented ID-less-prose fallback (paragraphItems), which treats every blank-line-separated paragraph in the file as its own new item, ID assigned positionally.

Net effect: a single legitimate multi-paragraph entry silently explodes into N fake sequential entries (e.g. one ## T0: entry with 7 paragraphs becomes items T0 through T6), and the fragmentation compounds any time a new paragraph is appended (a follow-up note, a later addendum) — each one becomes yet another fake entry. This lands directly in PRINCIPAL_TELOS.md, which is @-imported into every session. Reproduced on this install: a real 7-paragraph TRAUMAS.md entry and a 2-paragraph WRONG.md entry, both fragmented exactly as described.

Repro

Create a TELOS split file (e.g. WRONG.md) with:

## W0: Some belief
First paragraph of the entry.

Second paragraph, still the same entry, just a line break.

Run GenerateTelosSummary.ts. Output renders two separate bullet items instead of one — the second paragraph gets no ID at all (prefix is empty for that file), just a bare - text line.

Suggested fix

Extend parseItems()'s regex set to also match H2 ## ID: text headings (mirroring the existing H3 case at line ~131), and treat everything up to the next ## heading as that item's full body instead of falling through per-paragraph. The H2-form should take the same "explicit IDs always win" precedence the H3/bullet forms already get over the prose fallback.

[danielmiessler]

Fixed in source. parseItems now consumes ## ID: title sections (heading plus body up to the next heading) when no bullet or H3 items matched, so multi-paragraph bodies stay attached to their entry instead of fragmenting into positional fakes. Existing formats are untouched. The clean repro made this easy to verify. Ships with the next release.