Summary

docs cat (plain, --json, and --plain) silently drops person smart chips from its output. A tagged person in a doc is a person element — a sibling of textRun inside paragraph.elements — and the text extractor only concatenates textRun.content, so the chip contributes nothing to the rendered text. The same applies to date chips and richLink (Drive file / calendar event) chips.

The result is silently misleading output: a table row like Owner: @Jane Doe renders as Owner: with an empty cell, with no indication anything was skipped. --json doesn't help — it returns the same lossy string wrapped in {"tab": {"id", "index", "title", "text"}}. Today the only way to see who is tagged is cat --raw plus a hand-rolled jq walk.

Notably the write side already has first-class chip support (docs insert-person, #638, --at anchors in #683) — so gog can insert a person chip it then cannot read back.

Repro (self-contained, uses gog itself)

DOC=$(gog docs create "chip read test" -j | jq -r .id)
gog docs write "$DOC" --replace "Reviewer: "
gog docs insert-person "$DOC" --email [email protected] --at "Reviewer: "

gog docs cat "$DOC"
# Output: "Reviewer: "            <-- chip dropped, no marker

gog docs cat "$DOC" --json | jq -r .tab.text
# Same lossy text; no person data anywhere in the JSON

gog docs cat "$DOC" --raw | jq '[.. | objects | select(has("person")) | .person.personProperties]'
# [{"name": "Sample Person", "email": "[email protected]"}]  <-- data exists

Observed on v0.33.0 (also v0.22.0).

Expected

Two parts, either would help; both would be ideal:

1. cat text rendering: render chips inline instead of skipping them, e.g.

   • person → @Jane Doe <[email protected]>
   • date → the chip's display text
   • richLink → [Title](url)

   (Possibly behind a --chips flag if changing the default output is a concern, though rendering something seems strictly better than an invisible drop.)

2. cat --json structured output: include the chips as data, e.g. a persons / chips array per tab alongside text, so scripting consumers don't need to fall back to --raw + a recursive jq descent (which on multi-tab docs also means downloading the entire doc, since --raw returns all tabs).

Related

• docs: native smart-chip insertion (Person, Dropdown, Date, File) #638 — write-side chip insertion (shipped); this issue is the read-side mirror
• docs find-replace --first / format --match corrupt text in paragraphs containing non-text inline elements (chips, images): client-side offset math ignores skipped element index width #818 — chips corrupting find-replace offset math (chips as an editing hazard; here they're an extraction gap)
• feat(docs raw): add --tab to traverse a specific tab; default returns tab 0 only #697 — raw tab traversal (the current workaround path)