[lt-zeeshan]

What this adds

A card in the chat. A kane-cli test Claude runs (run, testmd run, testrun run) is drawn in the conversation in place of its shell row:

• one line while it runs, behind a spinner: name, time, current step;
• then a bordered card with the mascot: the result, a failure's on / why / kind, or a suite's cells, counts and up to three failures;
• View steps opens the pane on that run; View evidence opens the run's evidence pack in kane-cli's viewer in one click;
• where the chat is printed text (terminal outside fullscreen) the card ends with the evidence path instead of buttons;
• a command that never became a run keeps its normal row.

Use-case detail. Press a use case on the Assurance tab: title, risk, what it owes to design and to run, kane-cli's reason and next command, and Close these gaps with Claude.

History tab. The last 30 finished runs in the project (it was 10, shown only as last run …). /kane history opens it.

Remote suites. testrun run --remote: the band says how many tests are on the grid and since when, and the pane gives the job and its link, instead of sitting on "planning".

A spinner leads the running step on the band and the card, and the mascot's eyes step every 120 ms.

Found in live takes, and fixed

• Saved tests that match a change equally are ordered by last result (failed first, then most recent), not by folder order.
• A run started outside the chat is named from its own process's command line (ps), so a test file shows its file name.
• The idle band counts each test once, by its latest result: a failure that has since passed is gone.
• Use-case cards show their title; a reason shared by every row is said once.
• A long test name in the offer card is cut to its width instead of wrapping under the button.

One thing to decide

View evidence starts kane-cli evidence serve <pack> and opens the link it prints. That is the one kane-cli command the mod now starts besides cover gaps; it still never starts a test. It also reads that one printed line, because the viewer has no structured output. With no pack or no viewer it falls back to asking Claude, as before. If the "mod never starts anything" rule should stay absolute, this is the part to change.

How it was checked

• claude plugin test kane-qe: 127 pass (81 before). claude plugin validate passes for the plugin and the marketplace.
• Live: a real Claude Code session in a fullscreen terminal, a demo shop project with a real requirement store, and real kane-cli runs. guide/run_tour_v3.sh records that session; guide/make_promo.py cuts a captioned promo from a take.
• Not checked live: remote suites (written from kane-cli's docs/remote-execution.md, tests only); the browser tab opening after View evidence (the viewer start was seen, the tab was not recorded); the card's evidence-path form outside fullscreen; the desktop app.

guide/v2/design-coverage.md has the detail. The version is left at 2.0.0 for whoever releases.

🤖 Generated with Claude Code