Workflow startup greeting should render client-side, not round-trip through Claude

[jeremyeder]

Jira: RHOAIENG-51904

Problem

When a user selects a workflow (e.g., CLAUDE.md Generator), the welcome/greeting message takes 20+ seconds to appear in the chat UI. This is because the startupPrompt from ambient.json is sent through the full LLM round-trip:

runner → backend API → runner /agui/run → Claude SDK → MCP server startup (~14s) → Anthropic API (~3s) → AG-UI events → frontend

The greeting is a static message defined in the workflow's ambient.json. It doesn't need LLM processing.

Root Cause

_trigger_workflow_greeting in ambient_runner/endpoints/workflow.py sends the startupPrompt as a hidden user message to /agui/run, which spins up a full Claude session (including connecting all MCP servers) just to echo back a predetermined message.

Additional bug found during investigation

_trigger_workflow_greeting used path.split("/")[-1] to derive the workflow directory, while clone_workflow_at_runtime uses the repo name (git_url.split("/")[-1].removesuffix(".git")). These don't match, causing the greeting to silently fail — it can't find ambient.json and skips the greeting entirely.

Proposed Solution

1. Frontend renders the greeting directly: When a workflow is selected, the frontend reads the startupPrompt from the workflow metadata endpoint (/workflow/metadata) and renders it as a system/assistant message in the chat UI immediately — no Claude round-trip needed.

2. Claude is only invoked when the user sends a real message (e.g., types "1" to select PROJECT/REPO).

3. Remove _trigger_workflow_greeting from the runner or keep it as a fallback only.

Impact

• Every workflow activation has a 20+ second delay before the user sees anything
• Most of that time is MCP server cold starts (google-workspace downloads 80 Python packages via uv on every session start)
• The greeting itself is static text that could render in <100ms

Time Breakdown (from debug logs)

Phase	Duration
mcp-atlassian connect	~1s
webfetch connect (downloads cpython + deps via uv)	~4s
google-workspace connect (downloads 80 packages via uv)	~9s
Anthropic API call	~3s
Total	~20s

Steps to Reproduce

1. Create a session in the UI
2. Wait for MCP servers to show as connected
3. Click a workflow (e.g., CLAUDE.md Generator)
4. Observe 20+ second delay before greeting appears

[github-actions]

Amber Change Summary

The following files were modified:

 components/backend/handlers/sessions.go            | 35 +++++------
 .../components/welcome-experience.tsx              | 22 +++++++
 .../[sessionName]/hooks/use-workflow-management.ts |  9 ++-
 .../[name]/sessions/[sessionName]/lib/types.ts     |  1 +
 .../[name]/sessions/[sessionName]/page.tsx         |  9 ++-
 components/frontend/src/services/api/workflows.ts  |  1 +
 .../ambient_runner/endpoints/workflow.py           | 67 +---------------------
 7 files changed, 59 insertions(+), 85 deletions(-)

Next Steps:

• Review that changes match the issue description
• Verify no scope creep or unintended modifications
• A PR will be created shortly for formal review

---

🔍 View AI decision process (logs available for 90 days)

[github-actions]

🤖 Amber has created a pull request to address this issue: #780

The changes are ready for review. All automated checks will run on the PR.

---

🔍 View AI decision process (logs available for 90 days)