test(cockpit-chat-debug): aimock e2e suite for the c-debug devtools dock

[blove]

What changed

cockpit/chat/debug was the last chat capability with no aimock-replay e2e. It had none because the demo used to mount <chat-debug> alone — a devtools dock with no composer, so nothing could produce a run and the Timeline tab sat on its empty state forever. e2e/README.md documented that as a permanent gap.

The docs-audit work closed the gap in the demo: it now mounts <chat> beside <chat-debug> (both carrying the main slot attribute, since example-chat-layout only projects [main] and [sidebar]). This PR gives the capability the suite it never had, wired exactly like its siblings.

The suite

cockpit/chat/debug/angular/e2e/ — playwright.config.ts, global-setup-impl.ts, tsconfig.json, fixtures/c-debug.json, c-debug.spec.ts — plus an e2e target in project.json. That target is also what puts the cap in CI: scripts/cockpit-matrix.mjs derives the cockpit-e2e matrix from targets.e2e, so no workflow edit is needed (scripts/generate-aimock-scaffold.ts was not used — its ci.yml insertion predates that dispatcher).

Four specs:

1. the Timeline tab shows its empty state before any run;
2. a run sent through the composer fills it with the pipeline's checkpoint rows;
3. selecting a checkpoint diffs that step of the run (+ messages, read off the checkpoint the server persisted);
4. the State tab swaps in the live state inspector.

No assertion reads model prose. The checkpoint labels are structural — toDebugCheckpoint reads state.next[0] — so spec 2 asserts the exact row list (Step 1, generate_title, summarize, process, generate, __start__), newest first. That row set is precisely what this capability exists to demonstrate, so a node added to or dropped from the pipeline should fail here and be re-stated rather than silently pass.

The fixture

Three entries, one per LLM call the graph makes in a turn (generate, summarize, generate_title), discriminated by systemMessage because all three carry the same user message. The hasToolResult ordering rule does not apply here: the c-debug graph binds no tools.

Cross-references

cockpit/render/README.md and scripts/rerecord-all-aimock.sh both named c-debug as a documented no-e2e cap. The re-record script is the load-bearing one: it discovers caps by walking */e2e/fixtures/*.json, so c-debug is picked up automatically now, and an unregistered cap defaults to the prompt "Hello" — a re-record would have quietly replaced the fixture with one whose user message no spec sends. Its real prompt is now registered.

Verification

• npx nx build cockpit-chat-debug-angular — green.
• npx nx e2e cockpit-chat-debug-angular — 4 passed, run three times (--skip-nx-cache each time): 32.3s / 13.5s / 15.2s. Deterministic.
• npx nx smoke cockpit-chat-debug-angular — green.
• Mutation check: removing <chat> from the demo template makes the three run-driven specs fail on the missing composer while the empty-state spec still passes — exactly the state the deleted README described. Restored after.
• npx tsc -p cockpit/chat/debug/angular/e2e/tsconfig.json --noEmit — only the pre-existing baseUrl deprecation, identical to the tool-calls sibling.

Notes

• The fixture was authored, not recorded live. No OPENAI_API_KEY is available in this environment, so AIMOCK_MODE=record could not run. What is real is the run: the fixture drives the actual c-debug graph on a real langgraph dev server through the real adapter, all three of its LLM calls match and return 200, and every assertion is on structure the server produced (checkpoint rows, persisted checkpoint values), never on the fixture's own prose. Re-recording against the live provider with bash scripts/record-aimock-cap.sh c-debug "What is a jet bridge?" would swap the three response bodies for real ones without touching a single assertion.
• The State tab renders {} on this cap, so spec 4 asserts that the inspector is mounted and bound to the agent rather than asserting content. agent.state() is the LangGraph values bag with messages projected out into the transcript, and c-debug is a plain MessagesState graph, so there is nothing else in it. The tab is not broken; the graph simply carries no state beyond its messages. Giving it one (a steps list the pipeline appends to, say) would make the State half of the dock demonstrate something and is a small change — but it rewrites what the docs page teaches, so it is left as a call for the coordinator. Spec 3 is where "reflects the run" is actually asserted, against the checkpoint diff.
• apps/website/content/docs/chat/components/chat-debug.mdx needed no correction: it already describes the composer-driven flow and never claimed the demo lacked one or lacked e2e.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
threadplane	Ready	Preview	Sep 8, 2026 3:27pm UTC

[github-actions]

Automated approval: this PR received an intelligent (AI) code review. See the review comments on this PR.

[github-actions]

Claude finished @blove's task in 0s —— View job

---

I'll analyze this and get back to you.

[github-actions]

Claude finished @blove's task in 0s —— View job

---

I'll analyze this and get back to you.