# Featured: a real bug fix across surfaces `workflow.mp4` / `workflow.gif` follow one disposable **Fix task CLI status filtering** conversation using GPT-6.1-Sol. 1. `t open demo 1`; ask the agent to reproduce and fix `--status` filtering. 2. The agent runs three tests (two fail), edits `tasks.py`, adds two regression tests, and reruns the suite (all five pass). The real CLI prints only open tasks. 3. `t beam demo 1 --host mini --detach`; run the same five tests and CLI on mini; `t beam demo 1 --from mini --detach` returns the same task. 4. `t app demo 1 --reuse-window --no-preview --no-plan`; `t ls demo` verifies the same model, active context, task title, and desktop owner. 5. Archive only the demo chat, `t app pull demo 1 --thread `, end that demo CLI, then `t resume demo 1`. Ask the resumed agent to verify the same fix and tests. The `t app` invocation is the configurable desktop-handoff form; ordinary use is `t open demo 1 --app`. Window reuse avoids creating another app window during capture. The resume picker selects the named coding task when multiple saved chats exist. The explicit `t app pull --thread` selects this demo conversation because the fixture also contains the earlier checklist chat; `t open demo 1 --cli` is the ordinary return command for an unambiguous worktree. The before-state source and actual agent-produced patch are in [`scripts/demo/fixture`](../../scripts/demo/fixture/README.md). The fixture was independently replayed: two failures before the patch, five passing tests after it. `t beam` checkpoints the working edits into a local WIP commit when transferring. Captured October 11, 2026. The task ID was checked against the copied rollout on mini. The desktop's loaded-thread writer ownership was independently checked. The resumed terminal contains the same code-fix conversation. **Limitations:** mini's Codex is not logged in. The task, transcript, and code really moved to mini and back; tests ran there, but no remote model turn ran. The app window was not filmed. Desktop archival and ending the returned demo CLI happened between chapters. Terminal output is original PTY output. The video adds editorial chapter labels, joins captures in order, and removes idle gaps and repeated output. The featured cut is 66 seconds, including a two-second editorial title card. `render-cover.py` generates that card and the matching video poster. Reading sections play at normal speed. Session setup plays at 2×, marked on screen. The desktop-to-CLI return occurs between chapters. Setup retries are omitted; the final transfer retry follows the initial WIP checkpoint. This is an edited demonstration, not an uninterrupted screen recording. No command results are invented. The replay preserves ANSI foreground/background colors, bold, reverse video, and underlines. A presentation theme adds cyan prompts, green additions/passing results, and red removals/failures to monochrome source text; these accents are editorial styling, not a claim about the original CLI theme. The fixture uses isolated configuration, repositories, and tmux sockets on both hosts, a demo-scoped SSH wrapper to pass that environment, and a temporary Codex `--no-daemon` launcher. Repository paths, host address, and task IDs are redacted. The private fixture configuration and authentication material are not included. Capture actual commands using `scripts/demo/capture-step.py` with a private state file and output directory. `--select-after` presses Enter once for a previously observed picker selection; `--detach` detaches the tmux client. Inspect the real outputs and verify ownership before sanitizing sources for the repository. Regenerate offline (Python 3, Pillow, pyte, ffmpeg): ```sh python3 -m venv /tmp/t-media-venv /tmp/t-media-venv/bin/pip install Pillow pyte /tmp/t-media-venv/bin/python scripts/demo/render-flow.py ``` `workflow.json` defines chapter sources, trims, evidence, and limitations. Related work: [site #5](https://github.com/agenthangar/agenthangar.github.io/issues/5) and [t #107](https://github.com/agenthangar/t/issues/107). # Featured: RuntimeBrief on iPhone `runtimebrief.mp4` (75 seconds, 1920×1152, 60 fps) and `runtimebrief-poster.jpg` show the RuntimeBrief iPhone app's offline demo. Every project, commit, session, and reply in it is the app's fictional demo data; no Mac or real project is involved. 1. The portfolio: Claude Code, Codex, and Cursor work across projects in one list. 2. Sample Tracker: the analyst update, agent sessions from all three vendors, recent commits, and Xcode, TestFlight, and App Store status. 3. Siri: "What needs attention in RuntimeBrief?" and "Ask RuntimeBrief what's the status of Sample Tracker". 4. "Start Codex for Sample Tracker from RuntimeBrief": the task, Siri's confirmation, and the new Codex task in the app. 5. "Tell the Sample Tracker agent in RuntimeBrief": the follow-up appears in the Codex conversation. **Capture.** App footage is one continuous screen recording of build 1.0.3 (44) on a Revyl cloud iPhone 17 Pro Max (iOS 27), captured October 14, 2026. The edit removes idle time between taps, plays scrolling at 1.6×, and joins cuts on identical screens. The recording's clock and "no service" indicator are replaced by a 9:41 status bar. Tap circles, amber outlines, the device frame, captions, title and end cards are editorial. **Siri screens are recreations.** Siri can't be recorded on a cloud device, so its orb, request bubbles, and cards are drawn over the app footage, following Siri's layout on a physical iPhone. The replies use the app's own reply formats with the demo projects' facts. Demo mode prefixes Siri replies with "Fictional demo"; the video omits that prefix, and the Sample Tracker status answer is written in the analysis format rather than quoting the demo's own analysis text. **Music** is an original track synthesized for this video; it contains no samples. Agent logos follow [`assets/agent-logos`](../agent-logos/README.md) and are unchanged. The edit was made with Remotion; its source is kept in the [runtimebrief repository](https://github.com/agenthangar/runtimebrief) under `docs/demo-video`.