Event-Driven Requirements Review Agent

Lesson 5 (event-driven loop) companion repo for Loop Engineering with LangChain Deep Agents.

A Requirements Review Agent (built with LangChain / deepagents) that runs behind two real triggers instead of a notebook cell: a Slack bot you can message directly, and a Jira/Confluence webhook triggered by a real button on a ticket or a page. The agent itself never changes between the two — only how it gets invoked, and where the requirement document comes from.

Do Slack first — it's the simpler setup (no public URL needed). Jira/Confluence is the second section.

Want to adapt this to your own tools instead of Slack/Jira/Confluence, or change how the agent decides it's done? See EXTENDING.md — it's the design spec this repo was built from, written to be handed directly to a coding agent.

---

Prerequisites

• Python 3.12+
• An OpenAI API key (the agent makes live model calls)
• A Slack workspace where you can install an app (for the Slack section)
• An Atlassian (Jira + Confluence) Cloud site where you can create API tokens and Automation rules (for the Jira/Confluence section)

git clone <this-repo-url>
cd loop-engineering-event-driven
python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate
pip install -r requirements.txt
cp .env.example .env

You'll fill in .env as you go through each section below.

---

Part 1: Slack bot

What you're running

slack_bot.py — DM it directly, or @mention it in a channel it's been invited to, with a requirement document as the message. It replies in the same thread with the agent's analysis.

1. Set your OpenAI key

In .env:

OPENAI_API_KEY=sk-...

2. Create the Slack App

1. Go to api.slack.com/apps → Create New App → From scratch. Name it, pick your workspace.

2. OAuth & Permissions (left sidebar) → Scopes → Bot Token Scopes → add:

   • app_mentions:read
   • chat:write
   • im:history
   • channels:history
   • groups:history (needed for private channels, and for the bot to read a thread's parent message when it's @mentioned with little/no text)

3. Socket Mode (left sidebar) → turn it on. This generates an App-Level Token (xapp-...) — copy it.

4. Event Subscriptions (left sidebar) → turn it on → under Subscribe to bot events, add:

   • app_mention
   • message.im

   Both the scope and the event subscription are required — a scope only grants permission, the event subscription is what actually makes Slack push that event to you.

5. App Home (left sidebar) → turn on the Messages Tab. Without this, Slack silently drops DMs to the bot before your code ever sees them.

6. Install App (left sidebar) → Install to Workspace → Allow. This generates the Bot Token (xoxb-...) — copy it.

3. Set the Slack tokens

In .env:

SLACK_BOT_TOKEN=xoxb-...
SLACK_APP_TOKEN=xapp-...

4. Run it

python slack_bot.py

Leave this running in its own terminal tab.

5. Try it

• DM: open a direct message with the bot, paste a requirement document, send.
• Channel: invite the bot to a channel (/invite @YourBotName), post a requirement document, then @mention the bot underneath it (or in the same message).
• Short mention in a thread: post a requirement as one message, then reply in that thread with just @YourBotName — it infers the document from the thread's parent message.

If DMs don't seem to work even after this setup, it's almost always a desktop-client cache issue — reload the Slack app (Cmd+R / Ctrl+R).

---

Part 2: Jira + Confluence webhook

What you're running

jira_webhook.py — a small server with two routes. It needs to be publicly reachable, because (unlike Slack) Jira and Confluence can only call out to a URL you give them — they have no "stay connected and listen" mode. That's what ngrok is for below.

• Confluence (primary): a button on the requirement page itself. Click it any time — before a ticket even exists — and the review lands as a comment on the page.
• Jira (secondary): a button on a ticket. Click it to re-check the requirement linked to that specific ticket. Writes back to both the ticket and the linked Confluence page.

1. Get Atlassian API credentials

1. Go to id.atlassian.com → Account Settings → Security → API tokens → Create API token.
2. In .env:

   ATLASSIAN_EMAIL=you@example.com
   ATLASSIAN_API_TOKEN=<the token you just created>
   ATLASSIAN_SITE_URL=https://your-site.atlassian.net
   ATLASSIAN_WEBHOOK_SECRET=<any random string you choose>
   

   ATLASSIAN_WEBHOOK_SECRET isn't an Atlassian value — you make it up. It's the shared secret the Automation rules will send back to prove the request is really coming from your Jira/Confluence site.

2. Run the webhook receiver

python jira_webhook.py

Leave this running in its own terminal tab. It listens on localhost:5001.

3. Expose it publicly with ngrok

In a third terminal tab:

ngrok http 5001

Copy the https://...ngrok-free.app (or .dev) URL it prints. This stays the same across restarts as long as you don't recreate it — ngrok gives every free account one permanent "dev domain."

4. Link a Confluence page to a Jira ticket

Pick (or create) a Jira ticket and a Confluence page holding the actual requirement text. Link them natively: from the Jira ticket, Link → Link to Confluence page (or the equivalent from the Confluence side). The webhook finds the linked page through this real link, not a custom field.

5. Create the Jira Automation rule

1. Open the ticket → ⋯ or the ⚡ Automation icon in the issue panel → (or Project settings → Automation) → Create rule.
2. Trigger: search for and select Manual trigger.
3. Action → Send web request:
   • URL: <your-ngrok-url>/jira/webhook/analyze
   • Method: POST
   • Body: Custom data —

     {"issueKey": "{{issue.key}}"}

   • Headers: X-Webhook-Secret = the same value as ATLASSIAN_WEBHOOK_SECRET in your .env
4. Name the rule, Save and enable.

6. Create the Confluence Automation rule

1. Go to the Confluence space (not the Jira project) → the space's "⋯" menu → Automation → Create flow.
2. Trigger: Manual trigger from page.
3. Action → Send web request:
   • URL: <your-ngrok-url>/confluence/webhook/analyze
   • Method: POST
   • Body: Custom data —

     {"pageId": "{{page.id}}"}

   • Headers: X-Webhook-Secret = same value as above
4. Save and enable.

7. Try it

• Confluence: open the requirement page → hover its title in the left sidebar's Content list → "⋯" → Automate → click your rule. (Atlassian recently moved this out of the page header — if you don't see it there, this sidebar path is the current location.) A comment appears on the page within a minute or two.
• Jira: open the ticket → the ⚡ icon in the issue panel → your rule's name. Two comments appear within a minute or two: one on the ticket, one on the linked Confluence page.

Both responses come back fast (under a second) even though the actual review takes longer — the server acknowledges the click immediately and does the real work in the background, so neither Automation rule times out waiting.

---

Troubleshooting

Symptom	Likely cause
Slack DMs don't arrive	Messages Tab not enabled under App Home, or message.im event subscription missing (scope alone isn't enough)
Slack says "sending messages has been turned off" after you just enabled it	Desktop client cache — reload (Cmd+R / Ctrl+R)
Nothing prints when you start a bot	stdout is buffered when not attached to a terminal directly — run with python -u if redirecting output to a file
Jira/Confluence button click does nothing	Check the ngrok URL in the Automation rule still matches what ngrok http 5001 is currently printing — it changes if you recreate the tunnel
Confluence trigger button not where you expect	It's in the left sidebar's content-item "⋯" → Automate, not the page header's "⋯" menu
Webhook returns 401	X-Webhook-Secret header value doesn't match ATLASSIAN_WEBHOOK_SECRET in .env

---

Files

• slack_bot.py — the Slack trigger (Socket Mode, no public URL needed).
• jira_webhook.py — the Jira + Confluence trigger (/jira/webhook/analyze and /confluence/webhook/analyze), needs a public URL via ngrok.
• deepagent_helpers.py — the agent itself (build_review_agent, run_verification_loop_live), unchanged by either trigger.
• workspace/memories/AGENTS.md — the agent's instructions, loaded fresh into every run.
• jira_tickets.json — sample tickets for local testing without any real Jira/Confluence account (not used by the two triggers above, which always pass a real document in, but required for deepagent_helpers.py to import).
• EXTENDING.md — the design spec behind this repo. Use it as a starting point if you want to adapt the agent to a different trigger (a GitHub webhook, a cron job, a different chat tool) or hand it to a coding agent to build the next one for you.