Generate the hero topology GIF from a real Agent Channel run #57

New issue

Closed

opened 2026-06-04 12:02:25 +00:00 by coilysiren · 0 comments

coilysiren commented

2026-06-04 12:02:25 +00:00

Owner

Generate the README hero topology GIF from a real Agent Channel run instead of synthetic fixture spans.

The hero GIF was rendered from a hand-tuned synthetic fixture. o2r already has a corpus of real multi-agent runs - the Agent Channels exposed by the o2r CLI - so the hero should be drawn from one of those, which is both more honest and a better demonstration of "agent activity as OTel spans."

What lands

channels.py adapter: projects one channel's event log onto the topology span shape (channel = hub, each posting agent = a leaf, events ordered by created_at). Filters state roster heartbeats to the substantive hops. Hard scrub gate (assert_clean) so no tailnet IP / host FQDN can reach an on-screen label - this repo is public.
make channel-capture: pulls a live channel via the o2r CLI, sanitizes, freezes the spans to arize_phoenix/tests/fixtures/real_runs/ (we freeze sanitized spans, never raw events).
make gif-real-update: renders assets/real-session-topology.gif from the frozen fixture - deterministic, the same way the synthetic baseline is.
Hero run is DWUW: four agents coordinating a cross-node k3s flannel fix through one relay.

Shared renderer changes

The label/sidebar/duration fixes the real run needed touch the shared renderer (bigger canvas, n=4 diagonal-ellipse layout, drop the agent x -> y: log prefix, human-readable duration), so the synthetic byte-exact baseline is re-rendered too via the Linux docker path.

README hero + docs/FEATURES.md updated to describe the real-run pipeline.

Generate the README hero topology GIF from a **real** Agent Channel run instead of synthetic fixture spans. The hero GIF was rendered from a hand-tuned synthetic fixture. o2r already has a corpus of real multi-agent runs - the Agent Channels exposed by the o2r CLI - so the hero should be drawn from one of those, which is both more honest and a better demonstration of "agent activity as OTel spans." ## What lands - **`channels.py`** adapter: projects one channel's event log onto the topology span shape (channel = hub, each posting agent = a leaf, events ordered by `created_at`). Filters `state` roster heartbeats to the substantive hops. Hard scrub gate (`assert_clean`) so no tailnet IP / host FQDN can reach an on-screen label - this repo is public. - **`make channel-capture`**: pulls a live channel via the o2r CLI, sanitizes, freezes the spans to `arize_phoenix/tests/fixtures/real_runs/` (we freeze sanitized spans, never raw events). - **`make gif-real-update`**: renders `assets/real-session-topology.gif` from the frozen fixture - deterministic, the same way the synthetic baseline is. - Hero run is `DWUW`: four agents coordinating a cross-node k3s flannel fix through one relay. ## Shared renderer changes The label/sidebar/duration fixes the real run needed touch the shared renderer (bigger canvas, n=4 diagonal-ellipse layout, drop the `agent x -> y:` log prefix, human-readable duration), so the synthetic byte-exact baseline is re-rendered too via the Linux docker path. README hero + `docs/FEATURES.md` updated to describe the real-run pipeline.

coilysiren referenced this issue from a commit

2026-06-04 18:01:06 +00:00

feat(viz): render hero topology GIF from a real Agent Channel run

coilysiren closed this issue

2026-06-04 18:01:06 +00:00

Commenting is not possible because the repository is archived.