[crisperity]

What

• Overview now starts with 1-2 concept cards (title, before, after, why). Text is generated first, then one picture per concept, drawn in its own worker pool. If a picture fails, the text stays. Old walkthroughs without concepts render as before.
• Thread answers can define domain terms. The agent calls a new define_terms tool (max 5, only terms that occur in the answer). Terms are stored with the turn and shown as a dotted underline with a popover (hover, keyboard focus, long-press on touch).
• Plugin version 1.3.0 -> 1.4.0.

Notes

• The concepts step costs one extra LLM call per run, including docs-only changes.
• The old overview graph still renders under the concepts.
• test_cw_ask.py::test_full_turn_is_wrapped_in_bwrap_when_available fails inside the sandbox, unrelated.

Not checked

Nothing was run in a real browser: concept cards, mobile layout, dark mode, popover placement and touch long-press. Real model output for concepts and terms is also untested.