Files
brain-of-reese/.agents/phases/todo/182_verbatim_show_text
ducoterra 76887639f3
Build and Push Containers / build-and-push-app (push) Successful in 1m49s
Build and Push Containers / build-and-push-db (push) Successful in 13s
docs(phases): add phase 182 (verbatim file display)
Answers to "show me my pi.dev config" re-type file contents from context,
so the code block in the answer can differ from the indexed file by a
character and cannot be trusted for copy-paste — the owner opens the
source bubble and ignores the answer.

Phase 182 makes the server the only party that supplies bytes:

- show_text(path, offset?, limit?), the display sibling of show_image,
  slices documents.content and hands the same bytes back to the model
  with a "do not retype" instruction
- inline ```bor source/path 12-40 locator fences place a resolved block
  anywhere in the prose; the model's fence body is never painted, and an
  unresolvable locator says so instead of showing guessed text
- the bytes ride inline in the done frame's additive answer_text array
  and persist in the saved chat, so restore and the shared page match
- line numbers are always the server's resolved extent (overshoot
  clamps, past-the-end refuses to the model), which removes the
  "did the model invent those numbers?" problem entirely
- a Copy button on every answer code block, a VERBATIM badge only on
  server-resolved ones
- gated by BOR_ANSWER_TEXT with per-block/per-answer budgets; with the
  flag off every frame, record and byte budget is exactly pre-phase

Overview plus eight tasks, each with its own test gate; the phase E2E
runs in isolation and asserts the clipboard content equals the stored
bytes. Run 180 first — both touch markdown.js.

LLM-Generated: true
2026-10-14 10:33:06 -04:00
..