Conversation Launcher

What the source does

• FilesystemNone observedNo reads or writes in the source
• NetworkNone observedNo outbound calls in the source
• ShellNone observedNo commands run in the source
• DependenciesPinnedNothing floats
• SecretsNone foundNo keys in source

Five audit layers

Gates applied: no behavioural pass.

assistant/CLAUDE.md

assistant/src/approvals/CLAUDE.md

assistant/src/notifications/CLAUDE.md

assistant/src/runtime/CLAUDE.md

gateway/CLAUDE.md

Audit 0.4.1 · source sha 19bc86343b63 · 14 Oct 2026Report an issue or request a re-scan

• Publishervellum-ai GitHub organizationOwner of the repository this skill is read from
• Source foldervellum-ai/vellum-assistant/skills/conversation-launcherThe exact folder we audited
• LicenceMITAs the repository states it
• Commit pinned for this verdict19bc86343b63A new audit runs when the source changes
• Repository statusNot archivedLast push 1 day ago
• Sandboxed runNot runSkills are read, not executed; layer L4 is skipped

Its description: Offer the user several spin-off conversations as clickable buttons on a single persistent card. Each click spawns a fresh seeded conversation in the sidebar; the user keeps their place in the current conversation. Use…

---
name: conversation-launcher
description: Offer the user several spin-off conversations as clickable buttons on a single persistent card. Each click spawns a fresh seeded conversation in the sidebar; the user keeps their place in the current conversation. Use when you want to branch into N focused threads (research directions, draft choices, pending replies, triage of N items) without losing the current context. Not for single-destination pivots — just reply inline.
compatibility: Designed for Vellum personal assistants
metadata:
  emoji: "🧭"
  vellum:
    category: "productivity"
    display-name: "Conversation Launcher"
---

Use this skill when you want to offer the user several spin-off conversations from the current one. You render **one** persistent card. Each button on the card spawns its own seeded conversation in the sidebar. The user can click multiple buttons without losing their place — the origin conversation (this one) keeps focus.

## When this fits

- Research branches — "here are three angles to pursue"
- Draft choices — "here are the reply drafts I could write"
- Triaging N items — "here are the five threads with pending replies"
- Pending-reply fan-out — each sender gets their own drafting conversation

## When this does NOT fit

- Single-destination pivots — if there's one obvious next conversation, just reply inline or navigate there directly. One button is not a menu.
- Options that share context and should stay in one thread — keep them here.
- Inline Q&A the user can skim in place — answer; don't fan out.

## How to render

Emit exactly one `ui_show` call with a card shaped like this, then end your turn:

```json
{
  "surface_type": "card",
  "display": "inline",
  "persistent": true,
  "await_action": false,
  "data": {
    "title": "<framing headline>",
    "body": "<one short sentence framing the choice>"
  },
  "actions": [
    {
      "id": "opt-1",
      "label": "<short button label>",
      "style": "primary",
      "data": {
        "_action": "launch_conversation",
        "title": "<short conversation title>",
        "seedPrompt": "<full first-user-message seed>",
        "anchorMessageId": "<optional anchor message id from this conversation>"
      }
    },
    {
      "id": "opt-2",
      "label": "<short button label>",
      "style": "secondary",
      "data": {
        "_action": "launch_conversation",
        "title": "<short conversation title>",
        "seedPrompt": "<full first-user-message seed>"
      }
    }
  ]
}
```

Field notes:

- `persistent: true` keeps the card visible after a click so the user can fire more buttons.
- `await_action: false` lets the turn end without reserving the interactive-surface slot — the launcher dispatches its action directly, so blocking other surfaces is unnecessary.
- Each action's `data` must contain `_action: "launch_conversation"`, `title`, and `seedPrompt`. `anchorMessageId` is optional — include it when the spawned conversation should thread off a specific message in this one.
- `label` is the button text (short, ≤ 4 words, ≤ 30 chars). `title` is the new conversation's sidebar name (3–5 words, specific not generic). `seedPrompt` is the first user message of the new conversation — written in first-person as if the user typed it, with enough context that the new conversation can pick up without re-asking.
- Use `style: "primary"` for the recommended option (at most one), `style: "secondary"` for the rest.

## Do nothing else

After rendering the card, end your turn. The click pipeline handles everything:

- Do NOT run a follow-up step, issue any HTTP request, or write any file.
- Do NOT send a chat message after the card. The card is the response.
- Do NOT re-render the card on each click — it stays visible.

## UX contract the client enforces

- The card stays open and clickable for the lifetime of the live conversation — multiple buttons can be fired in succession without losing the user's place. If the conversation is closed and reopened later from history, the restored card reverts to single-click behavior (history rehydration drops the `persistent` flag — out of scope for this skill).
- Each `action.id` fires at most once per card lifetime; sibling actions remain clickable.
- Each spawned conversation inherits this conversation's guardian / trust context automatically.
- Sidebar focus stays on this conversation. The user sees new entries appear and can navigate in at their pace.

## Authoring tips

- Don't invent options. Only surface branches the user has actually discussed or implied.
- 2–5 buttons is the sweet spot. One option? Just reply inline. More than five? Trim or group.
- Keep the card body to one sentence. The buttons carry the payload; the body just frames the choice.
- Conversation titles should be highly concise, context-specific, and 3–5 words.

Excerpt, first 4,842 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.

git clone https://github.com/vellum-ai/vellum-assistant
git -C vellum-assistant checkout 19bc86343b63

Works in

1. 2026-10-14
   ✓SAFE B 89 / 100

   first audit · source 19bc86343b63

Questions about Conversation Launcher

• github.com/vellum-ai/vellum-assistant/tree/main/skills/conversat — the audited source at commit 19bc86343b63 and every finding aboveObserved · 14 Oct 2026

Spotted something wrong? Report a correction · Methodology