feat(agents): applications landing polish + chat dedup fix

[dmarticus]

Polish on top of #2700 — focused on the Applications landing IA, the Agent Builder dock's distinct identity, and a real bug in the chat preview's user-message dedup.

Stacked on top of posthog-code/m3a-final so the diff stays scoped. Merge order: #2700 → this.

Summary

Landing polish (commit 7a2094ed9)

• Single entry point to the Agent Builder. Dropped the contextual New agent / Edit with AI header CTA across every agents view; the always-on sparkles dock toggle is now the only entry point. Removes the "two near-identical gold buttons" confusion in the page header. Deleted EditWithAIButton + agentBuilderActions.
• Tinted the dock with a subtle amber accent (left border + faint header wash) so it reads as the meta-tool rather than another task chat. Mirrors the existing SparkleIcon accent.
• Reordered the Applications landing. Agents list moves to the top, sectioned LIVE vs DRAFTS (drafts dimmer, count in the section label, hidden when empty). Activity rollup hides when analytics.empty. Operational strip drops the live-count chip — the always-visible Live now panel below owns that signal — and keeps only the pending-approvals link, amber-tinted when non-zero.
• Tightened the Applications shell copy to mirror Scouts: "Deployed agents triggered by chat, Slack, webhooks, or cron — preview, approve, and edit with the Agent Builder." (single line at the current width).

Chat dedup fix (commit c5cda4811)

The optimistic-seed → SSE-echo dedup in createAgentChatMapper used strict equality on pendingOptimistic[0], which broke in three real failure modes — each one surfaced the same way: the user's just-sent bubble rendered twice.

Failure mode	Symptom	Fix
Whitespace asymmetry	Runner echoes back with trailing \n or padding around the context envelope; strict equality misses by one character.	Compare text.trim() to pending.trim().
Out-of-order echo	Rapid back-to-back sends land out of order; [0] check only inspects the head.	findIndex over the whole queue.
Runner re-emit	Runner emits the same user_message twice; second arrival has nothing in pendingOptimistic to swallow it.	Track every rendered user text in a seenUserTexts set; drop later duplicates.

Four new tests in sessionEventToAcp.test.ts covering each failure mode. 15/15 pass; pnpm --filter @posthog/ui typecheck clean.

Test plan

• Apps list looks right: LIVE/DRAFTS sections, drafts dimmer, hidden when empty.
• Open the dock; it has the amber accent and no longer competes with a page-header CTA.
• Pending-approvals link still routes to /code/agents/applications/approvals and tints amber when non-zero.
• Live now panel renders its own empty state when the fleet is quiet.
• Send a message in the Agent Builder dock as a fresh session — user bubble appears once (not twice).
• Send a message in a per-agent Chat preview tab — same.
• Send two messages back-to-back fast — both render exactly once each.

🤖 Generated with Claude Code

[github-actions]

React Doctor found no issues in the changed files. 🎉

_{Reviewed by React Doctor for commit c5cda48.}

[greptile-apps]

Prompt To Fix All With AI

Fix the following 2 code review issues. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 2
packages/ui/src/features/agent-applications/chat/sessionEventToAcp.test.ts:52-64
Per the repo's preference for parameterised tests, the first two new dedup tests probe the same behaviour (seed + echo → swallowed) with different string variants. Collapsing them into a single `it.each` table is more concise and makes adding more whitespace variants (e.g., leading spaces, both ends) trivial.

```suggestion
  it.each([
    ["exact match", "hello", "hello"],
    ["trailing newline", "hello", "hello\n"],
    ["leading spaces", "hello", "  hello"],
  ])(
    "swallows the echo of an optimistically-seeded message (%s)",
    (_, seeded, echoed) => {
      const mapper = createAgentChatMapper();
      mapper.seedUserMessage(seeded);
      expect(mapper.apply(ev("user_message", { text: echoed }))).toEqual([]);
    },
  );
```

### Issue 2 of 2
packages/ui/src/features/agent-applications/components/AgentApplicationsListView.tsx:261-265
The JSDoc says "only rendered when something is actually happening," but `OperationalStrip` is called unconditionally in the parent and renders even when `pendingCount === 0` (showing "0 pending approvals" as a navigation link). The comment should be updated to reflect actual behaviour.

```suggestion
/**
 * Operational counts strip — always rendered as a persistent navigation link
 * to the fleet approvals queue. Tinted amber when there are pending approvals,
 * neutral otherwise.
 */
```

_{Reviews (1): Last reviewed commit: "fix(agents): dedup user_message echoes r..." | Re-trigger Greptile}

[greptile-apps]

Per the repo's preference for parameterised tests, the first two new dedup tests probe the same behaviour (seed + echo → swallowed) with different string variants. Collapsing them into a single it.each table is more concise and makes adding more whitespace variants (e.g., leading spaces, both ends) trivial.

Suggested change

	it("swallows the echo of an optimistically-seeded message", () => {
	const mapper = createAgentChatMapper();
	mapper.seedUserMessage("hello");
	expect(mapper.apply(ev("user_message", { text: "hello" }))).toEqual([]);
	});
	it("swallows the echo even when the runner adds trailing whitespace", () => {
	const mapper = createAgentChatMapper();
	mapper.seedUserMessage("hello");
	// Runners commonly normalize by adding a trailing newline — that mustn't
	// break dedup or the user sees their bubble twice.
	expect(mapper.apply(ev("user_message", { text: "hello\n" }))).toEqual([]);
	});
	it.each([
	["exact match", "hello", "hello"],
	["trailing newline", "hello", "hello\n"],
	["leading spaces", "hello", " hello"],
	])(
	"swallows the echo of an optimistically-seeded message (%s)",
	(_, seeded, echoed) => {
	const mapper = createAgentChatMapper();
	mapper.seedUserMessage(seeded);
	expect(mapper.apply(ev("user_message", { text: echoed }))).toEqual([]);
	},
	);

Context Used: Do not attempt to comment on incorrect alphabetica... (source)

Prompt To Fix With AI

This is a comment left during a code review.
Path: packages/ui/src/features/agent-applications/chat/sessionEventToAcp.test.ts
Line: 52-64

Comment:
Per the repo's preference for parameterised tests, the first two new dedup tests probe the same behaviour (seed + echo → swallowed) with different string variants. Collapsing them into a single `it.each` table is more concise and makes adding more whitespace variants (e.g., leading spaces, both ends) trivial.

```suggestion
  it.each([
    ["exact match", "hello", "hello"],
    ["trailing newline", "hello", "hello\n"],
    ["leading spaces", "hello", "  hello"],
  ])(
    "swallows the echo of an optimistically-seeded message (%s)",
    (_, seeded, echoed) => {
      const mapper = createAgentChatMapper();
      mapper.seedUserMessage(seeded);
      expect(mapper.apply(ev("user_message", { text: echoed }))).toEqual([]);
    },
  );
```

**Context Used:** Do not attempt to comment on incorrect alphabetica... ([source](https://app.greptile.com/review/custom-context?memory=instruction-0))

How can I resolve this? If you propose a fix, please make it concise.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

[greptile-apps]

The JSDoc says "only rendered when something is actually happening," but OperationalStrip is called unconditionally in the parent and renders even when pendingCount === 0 (showing "0 pending approvals" as a navigation link). The comment should be updated to reflect actual behaviour.

Suggested change

	/**
	* Operational counts strip — restores the "live now / pending approvals"
	* signals the M7 analytics KPIs displaced. Live count anchors the live-now
	* panel below; pending links to the fleet approvals queue.
	* Operational counts strip — only rendered when something is actually
	* happening. Pending deep-links to the fleet approvals queue; live anchors the
	* live-now panel below.
	*/
	/**
	* Operational counts strip — always rendered as a persistent navigation link
	* to the fleet approvals queue. Tinted amber when there are pending approvals,
	* neutral otherwise.
	*/

Prompt To Fix With AI

This is a comment left during a code review.
Path: packages/ui/src/features/agent-applications/components/AgentApplicationsListView.tsx
Line: 261-265

Comment:
The JSDoc says "only rendered when something is actually happening," but `OperationalStrip` is called unconditionally in the parent and renders even when `pendingCount === 0` (showing "0 pending approvals" as a navigation link). The comment should be updated to reflect actual behaviour.

```suggestion
/**
 * Operational counts strip — always rendered as a persistent navigation link
 * to the fleet approvals queue. Tinted amber when there are pending approvals,
 * neutral otherwise.
 */
```

How can I resolve this? If you propose a fix, please make it concise.

[benjackwhite]

Personally i think this is a regresssion :( Having the clear prompt to push to the agent builder makes a lot of sense. Otherwise it isn't at all clear that this is even an option.

The sparkle button isn't obvious enough imo and i think this change isn't worth it

[benjackwhite]

I'll PR against it with what i think is preferable