From a7236050cf742ddbb82df181d6dbdcba776b8991 Mon Sep 17 00:00:00 2001
From: Miya <khaliqgant+miya@gmail.com>
Date: Fri, 11 Sep 2026 23:49:15 +0200
Subject: [PATCH] =?UTF-8?q?feat(examples):=20prospect=20demo=20flow=20?=
 =?UTF-8?q?=E2=80=94=20LLM=20to=20Slack=20to=20observer=20(#344)?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 examples/prospect-demo/README.md    | 64 +++++++++++++++++++++++++++++
 examples/prospect-demo/demo.flow.ts | 11 +++++
 2 files changed, 75 insertions(+)
 create mode 100644 examples/prospect-demo/README.md
 create mode 100644 examples/prospect-demo/demo.flow.ts

diff --git a/examples/prospect-demo/README.md b/examples/prospect-demo/README.md
new file mode 100644
index 000000000..21ec13af3
--- /dev/null
+++ b/examples/prospect-demo/README.md
@@ -0,0 +1,64 @@
+# Prospect demo
+
+Generate a short message with `f.llm`, post it to Slack's `#test` channel with
+`f.slack.post`, and open the observer URL printed by `flows run`.
+
+You need Node.js 22.18+, an installed and authenticated Claude CLI, a relayfile
+Slack mount connected to your workspace with permission to post in `#test`,
+and a Relaycast workspace key for the observer link. Create `#test` and invite
+the connected Slack bot before running the demo.
+
+Copy `demo.flow.ts` into a new directory, then run these commands there:
+
+```sh
+npm init -y
+npm pkg set type=module
+npm install relayflows @relayflows/surface
+cat > flows.json <<'JSON'
+{ "cli": "claude" }
+JSON
+```
+
+To use an authenticated Codex CLI instead, change `"claude"` to `"codex"` in
+`flows.json`. The LLM step uses that CLI's configured model and credentials.
+
+Point to your existing relayfile mount and configure the observer workspace:
+
+```sh
+export RELAYFILE_MOUNT_PATH='/absolute/path/to/your/relayfile-mount'
+export RELAYCAST_WORKSPACE_KEY='rk_live_REPLACE_WITH_YOUR_WORKSPACE_KEY'
+```
+
+The mount must contain a working `slack/` integration that confirms delivery.
+Setting `SLACK_BOT_TOKEN` alone is not enough: the current Slack helper requires
+relayfile writeback. See the [Slack helper documentation](../../docs/SLACK-HELPER.md).
+If `agent-relay` already has an active workspace saved locally, you can omit
+`RELAYCAST_WORKSPACE_KEY`; the CLI uses that workspace automatically.
+
+Check the flow and run it:
+
+```sh
+npx flows check demo.flow.ts
+npx flows run demo.flow.ts --local-agent --input '{}'
+```
+
+With `flows` on your PATH, the equivalent command is
+`flows run demo.flow.ts --local-agent --input '{}'`. The local worker runs the
+LLM call; the Slack helper waits for a delivery receipt before completing.
+Each new run posts a new message to `#test`.
+
+On success, look for the generated message in Slack and a terminal line like:
+
+```text
+Observer: https://agentrelay.com/observer?key=ot_live_...
+```
+
+Open that URL to view the configured Relaycast workspace. The CLI mints the
+observer token and prints the link; the flow needs no separate observer call.
+If the line is missing, check the workspace key, unset `FLOWS_NO_OBSERVER`, and
+look for an `[observer]` diagnostic. `flows observer` can mint a link separately.
+The link uses a scoped observer token, not the workspace key.
+
+For a local preview without sending a Slack message, prefix the run command
+with `RELAYFLOWS_SLACK_MOCK=1`. The LLM call still runs, but Slack delivery is
+replaced by a local receipt; this does not demonstrate live Slack delivery.
diff --git a/examples/prospect-demo/demo.flow.ts b/examples/prospect-demo/demo.flow.ts
new file mode 100644
index 000000000..5773c9c15
--- /dev/null
+++ b/examples/prospect-demo/demo.flow.ts
@@ -0,0 +1,11 @@
+import { flow } from '@relayflows/surface';
+
+export default flow('prospect-demo', async (f) => {
+  const message = await f.llm`Write one short, friendly Slack message introducing
+    Relayflows: a flow can generate a message with an LLM and post it to Slack.
+    Return only the message text. Do not use tools or mention anyone.`;
+
+  await f.slack.post('#test', message);
+  f.done('success');
+  // `flows run` prints the observer URL after the run summary when configured.
+});