fix(code): show user skills in / popup and drop the autofill

[richardsolomou]

Problem

In running tasks, the / popup only showed /good /bad /feedback until the agent's available_commands_update arrived. And / mid-prompt opened an inline autocomplete instead of the popover.

Changes

• Running tasks fall back to the trpc-fetched skills list while session commands are empty. Loader moved to PromptInput so session views populate it too.
• / opens the popover anywhere in the prompt (matching @ and #), instead of the inline ghost-text autofill. The autofill code stays in tree but is no longer wired up.
• An explicit empty available_commands_update from the agent now suppresses the draft-store fallback, so the agent stays authoritative.

How did you test this?

• pnpm --filter code typecheck
• pnpm --filter code exec vitest run src/renderer/features/message-editor/ (70 tests pass, 6 in the new getSuggestions.test.ts)

Publish to changelog?

no

[greptile-apps]

Prompt To Fix All With AI

Fix the following 1 code review issue. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 1
apps/code/src/renderer/features/message-editor/suggestions/getSuggestions.test.ts:89-148
**Prefer parameterised tests for the fallback/override scenarios**

Tests 2 and 3 share identical setup (`seedSessionContext(TASK_ID)`) and differ only in whether `seedSessionAvailableCommands` is called; likewise tests 3 and 5 both exercise the draft-store fallback path but from different context states. Expressing these as `it.each` suites would satisfy the team's parameterised-test preference and make adding future cases (e.g. empty `availableCommands` from agent) a one-liner rather than a new `it` block.

_{Reviews (1): Last reviewed commit: "fix(code): show user skills in running t..." | Re-trigger Greptile}

[adboio]

this is a regression, i'm gonna take a peek in case there's a smaller/simpler bug here we can fix!

[adboio]

@richardsolomou clarifying - mid-prompt you should still be able to see commands, but they just "autofill" instead of showing the whole popover like you'd see at the start of the task. is that also not working for you?

[richardsolomou]

@adboio that does work, but is that the intended way of doing it? It feels..off. why don't we just reuse the same approach? Similar to Slack mentions, you can @ someone at the beginning or the middle of the message and you have the same behavior

[greptile-apps]

Prompt To Fix All With AI

Fix the following 1 code review issue. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 1
apps/code/src/renderer/features/message-editor/suggestions/getSuggestions.ts:166-169
Draft fallback fires when agent sends an explicit empty command list. `getAvailableCommandsForTask` returns `[]` both when no `available_commands_update` has arrived yet and when the agent sends one with `availableCommands: []`. The `sessionCommands.length > 0` guard cannot distinguish these two cases, so if an agent deliberately clears its command list the UI will still show the draft-store fallback — contrary to the stated invariant that "agent-supplied commands remain authoritative once they arrive." Storing a sentinel (e.g. `null` for "not yet received" vs `[]` for "agent sent empty") would let the guard be precise.

```suggestion
  const sessionCommands = taskId ? getAvailableCommandsForTask(taskId) : null;
  const draftCommands = store.commands[sessionId] ?? [];
  // null  → agent hasn't reported yet, use draft fallback
  // []    → agent reported an empty list, respect it (no custom commands)
  // [...] → agent reported commands, use them
  const agentCommands = sessionCommands ?? draftCommands;
```

_{Reviews (2): Last reviewed commit: "fix(code): show user skills in running t..." | Re-trigger Greptile}