[pratikbuilds]

Add deterministic samples built only from the conversation contract: a short chat, a 300-message history, a tool storm with a batch, a failure and every approval outcome, generated images, a thread and a failed send. playTurn reveals a message's parts over time for streaming stories. Images are inline SVG, timestamps are fixed, and nothing reads the clock or Math.random.

Stacked on #116: review only this commit. Part of #94.

Notes for review

• Sample unit tests live beside the samples under stories/ and run with bun test stories/chat/samples; bun run test only covers src/.
• Streaming stories built on playTurn should start paused or play on a user action: the axe sweep waits for CSS animations, not timers.
• Follow-up: stories/chat/conversation.stories.tsx still has its own placeholder history; move it onto the samples when it first renders contract messages.

Verification

• This commit passes bun run typecheck and bun run lint on its own (the stack is checked commit by commit).
• At the top of the stack: bun run check, bun run build, bun run dep-guard, bun run contrast-test and bun run test:e2e (Playwright specs and the Storybook axe sweep in light and dark) pass.

Closes #96

[corbits-triage]

Triage: Pass

No changes needed from you. A maintainer will review this pull request.

Stacked on #116: only this pull request's own changes were checked.

Notes:

• added code matches a code pattern: stories/chat/samples/samples.stories.tsx:47 Invented fallback `?? 0`; handle the missing value instead.

[TheGreatAxios]

Automated multi-agent review: independent review + second-agent validation of every flag; only confirmed items listed. Everything below is minor and verified by running the helper.

• minor stories/chat/samples/play-turn.ts:49-60 — a queued call (status: "pending", no approval record) replays as running first, then regresses to pending: toolStages only short-circuits approval-requested and running, so bare queued traces fall through to the synthetic-running branch (verified repro). Add part.status === "pending" to the no-stage branch plus a regression test asserting a queued call never shows running before its final frame.
• minor stories/chat/samples/play-turn.ts:100 — turnFrames can emit a duplicate no-op frame after a tool batch whose last call contributes no stage (final running/approval-requested with no record makes started[last] equal the settled form; verified). Fix: skip a frame whose parts equal the previous frame's, or stop the batch loop one short.

[pratikbuilds]

Thanks for the review. Both points are fixed.

1. A queued call replayed as running first, then went back to pending (minor)

• The fix: a queued call (pending, no approval record) now shows as queued from the first frame, just like a waiting or running call.
• Test: "shows a queued call in its final form at once, never as running".

2. A tool batch could produce a duplicate frame (minor)

• The fix: the batch now steps through only the calls that actually change, one at a time, and leaves the rest in their final form. This also covers a stage-less call in the middle of a batch, and a batch where no call has a stage, which stopping the loop one short would have missed.
• Tests: "never repeats a frame after a batch call that has no stage", plus a check across every sample message.

Commit: fix(stories): replay queued and stage-less tool calls faithfully

[TheGreatAxios]

Review · Approve

Adds sample conversations and a play-turn helper for the chat stories.

Findings

No findings.

[pratikbuilds]

Round 3 fixes

A fresh review pass over the stack turned up the points below; each is fixed in a new commit at the end of this branch, and the stack is rebased onto it.

• 8eb4d54: The Samples caption described a replay no story performs; it now says what the Frames column counts.

bun run check and dep-guard pass on every branch of the stack, and the full e2e suite passes at the top of the stack (163/163, including the stories axe sweep). Each behaviour fix comes with a test that fails without it; the forwardRef tests can only guard, since the suite runs on React 19.