Channel CONTEXT.md: agent generation + inject into task prompts (local & cloud)

[raquelmsmith]

What

Builds out the per-channel CONTEXT.md (folder instructions) feature across three areas, for both local and cloud runs:

1. Generate CONTEXT.md with an agent

• The CONTEXT.md empty state gains a "Generate with agent" action with a Local/Cloud toggle (reusing FolderPicker / GitHubRepoPicker + useUserRepositoryIntegration).
• Generation runs as a normal task in the channel's repo that explores the code + PostHog data (flags, experiments, surveys, notebooks, insights, analytics) and publishes the file via the PostHog MCP (desktop-file-system-instructions-partial-update).
• The view tracks that task: a centered "Generating" + View task state while it runs, "Generate again" if it stops before publishing, and the rendered doc once published. Status is environment-aware (cloud cloudStatus/latest_run.status, local live session).

2. Channel CONTEXT.md injected into task prompts

• When a task is created from a channel, that channel's CONTEXT.md is appended to the agent's initial prompt as optional background — framed as reference material, not instructions. Works for local and cloud tasks.
• New-task input shows a dismissable Using: #channel CONTEXT.md chip (dismiss drops it from the prompt).

3. Conversation rendering

• The injected <channel_context> block renders as a clickable #channel CONTEXT.md tag instead of inline text; clicking opens the exact snapshot that was sent as a real file tab in the split (a new context panel-tab type), not a live re-fetch.

Shared, server-side generation tracking

• "Which task is generating this folder's CONTEXT.md" is stored server-side, keyed on the folder (project-shared, visible to all users), replacing fragile local electronStorage. New api-client get/setDesktopFolderGenerationTask + useFolderGenerationTask hook.

⚠️ Requires a backend change (separate PostHog cloud repo)

The shared generation indicator depends on new endpoints that must be added in the PostHog cloud repo:

GET /api/projects/{team_id}/desktop_file_system/{id}/context_generation/  -> { task_id: uuid|null }
PUT /api/projects/{team_id}/desktop_file_system/{id}/context_generation/  body { task_id: uuid|null }

plus auto-clear of the association when a new folder-instructions version is published.

Until that lands, the client no-ops gracefully: generation still runs and the file still renders on publish — only the project-shared in-progress indicator is dormant. (Everything else in this PR works without backend changes.)

Testing

• Full pnpm typecheck (22/22), pnpm lint, node scripts/check-host-boundaries.mjs — all green.
• Unit tests: prompt-builder.test.ts (channel-context block/text), channelContext.test.ts (conversation tag parser).
• Manual: local + cloud generate, channel-context injection + tag for both run modes.

---

Created with PostHog Code

[github-actions]

React Doctor could not complete this scan.

No React dependency found in /tmp/react-doctor-baseline-14U8ZC/package.json. Add "react" to dependencies (or peerDependencies) and re-run.

_{Reviewed by React Doctor for commit 2f60498.}

[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/core/src/editor/prompt-builder.test.ts:7-11
The null/empty cases for both `buildChannelContextText` and `buildChannelContextBlock` are tested with multiple assertions inside a single `it` block. The team rule is to always prefer parameterised tests (`it.each`), which would also give a clear label per failing input if one regresses.

```suggestion
describe("buildChannelContextText", () => {
  it.each([
    [undefined],
    ["   \n "],
  ] as const)("returns null for empty or whitespace content (%s)", (input) => {
    expect(buildChannelContextText(input)).toBeNull();
  });
```

### Issue 2 of 2
packages/core/src/editor/prompt-builder.test.ts:29-35
Same pattern — four inputs collapsed into one `it` block. Each would be a separate row in an `it.each` table, making it immediately obvious which specific input fails.

```suggestion
describe("buildChannelContextBlock", () => {
  it.each([
    [undefined],
    [null],
    [""],
    ["   \n  "],
  ] as const)("returns null for empty or whitespace content (%s)", (input) => {
    expect(buildChannelContextBlock(input)).toBeNull();
  });
```

_{Reviews (1): Last reviewed commit: "feat(canvas): cloud support for channel ..." | Re-trigger Greptile}

[greptile-apps]

The null/empty cases for both buildChannelContextText and buildChannelContextBlock are tested with multiple assertions inside a single it block. The team rule is to always prefer parameterised tests (it.each), which would also give a clear label per failing input if one regresses.

Suggested change

	describe("buildChannelContextText", () => {
	it("returns null for empty or whitespace content", () => {
	expect(buildChannelContextText(undefined)).toBeNull();
	expect(buildChannelContextText(" \n ")).toBeNull();
	});
	describe("buildChannelContextText", () => {
	it.each([
	[undefined],
	[" \n "],
	] as const)("returns null for empty or whitespace content (%s)", (input) => {
	expect(buildChannelContextText(input)).toBeNull();
	});

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/core/src/editor/prompt-builder.test.ts
Line: 7-11

Comment:
The null/empty cases for both `buildChannelContextText` and `buildChannelContextBlock` are tested with multiple assertions inside a single `it` block. The team rule is to always prefer parameterised tests (`it.each`), which would also give a clear label per failing input if one regresses.

```suggestion
describe("buildChannelContextText", () => {
  it.each([
    [undefined],
    ["   \n "],
  ] as const)("returns null for empty or whitespace content (%s)", (input) => {
    expect(buildChannelContextText(input)).toBeNull();
  });
```

**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]

Same pattern — four inputs collapsed into one it block. Each would be a separate row in an it.each table, making it immediately obvious which specific input fails.

Suggested change

	describe("buildChannelContextBlock", () => {
	it("returns null for empty or whitespace content", () => {
	expect(buildChannelContextBlock(undefined)).toBeNull();
	expect(buildChannelContextBlock(null)).toBeNull();
	expect(buildChannelContextBlock("")).toBeNull();
	expect(buildChannelContextBlock(" \n ")).toBeNull();
	});
	describe("buildChannelContextBlock", () => {
	it.each([
	[undefined],
	[null],
	[""],
	[" \n "],
	] as const)("returns null for empty or whitespace content (%s)", (input) => {
	expect(buildChannelContextBlock(input)).toBeNull();
	});

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/core/src/editor/prompt-builder.test.ts
Line: 29-35

Comment:
Same pattern — four inputs collapsed into one `it` block. Each would be a separate row in an `it.each` table, making it immediately obvious which specific input fails.

```suggestion
describe("buildChannelContextBlock", () => {
  it.each([
    [undefined],
    [null],
    [""],
    ["   \n  "],
  ] as const)("returns null for empty or whitespace content (%s)", (input) => {
    expect(buildChannelContextBlock(input)).toBeNull();
  });
```

**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!