Convert the single remaining TODO item (TODO.md L3) into phase
124_image_chips_and_answers (Protocol B append; numbered after 123).
Two workstreams in one phase:
- Bug: retrieved image docs never earn a source chip because the
phase-119 cited tier is read-docs-only and a model never reads an
image doc (its seed block already holds its entire content). Fix by
auto-citing suggested-tier image docs.
- Feature: new toggle-gated show_image agent tool (BOR_ANSWER_IMAGES,
default off) so the LLM can surface a supporting image in the answer;
relevance gate is the model's decision; one image per answer; text
answer always provided; done-frame answer_image + saved/shared
persistence + the answer figure.
7 tasks: 01 toggle, 02 tool, 03 prompt clause, 04 frame+persistence,
05 auto-cite bug fix, 06 UI answer figure, 07 mock trigger + E2E.
TODO.md cleared to a bare title (items now live in .agents/phases/).
LLM-Generated: true
ducoterra
7bd6335929