A Slack bot that answers engineer questions from your Confluence specs, runs live AI-assisted brainstorm sessions, edits existing spec pages section by section, shows a changelog of what changed in a spec over any time period, and logs all spec activity to a master audit trail.
Built with Claude (Anthropic), Recall.ai, Confluence, and Slack. Deployable on Railway in under an hour.
Engineers ask questions in Slack. SpecBot searches your Confluence spec pages, passes the relevant content to Claude, and returns a grounded answer with source links — no hallucination, no invention beyond what the spec says.
/specbot What is the error handling flow for the login feature?
@SpecBot Does the spec cover offline mode?
Give SpecBot a title and a one-line description. It generates a fully structured spec in your team's format and publishes it to Confluence immediately — no brainstorm session needed.
/specbot create Payments v2 | handles retry logic for failed transactions
/specbot create Offline Mode | allow users to read cached content without a network connection | format like Login Flow Spec
By default the format matches your gold standard spec (CONFLUENCE_GOLD_STANDARD_PAGE_ID). The optional format like <page name> segment uses any other existing spec page as the template instead.
Start a live session in Slack. Talk through the feature idea in the thread naturally. SpecBot builds a structured proposal in Confluence as you go, updating it every 60 seconds. When you're done, one command finalises and publishes it.
/specbot brainstorm
SpecBot joins your Google Meet via Recall.ai, listens to the conversation in real time, and writes the proposal directly to a Confluence draft page as you talk. When the call ends, the proposal is finalised automatically.
/specbot live https://meet.google.com/abc-defg-hij
Point SpecBot at an existing spec page when joining a call. Instead of creating a new proposal, it monitors the conversation and patches only the sections where the team makes a concrete decision — directly updating the live spec page every 60 seconds.
/specbot live https://meet.google.com/abc-defg-hij | Payments v2 Spec
SpecBot will tell you in Slack exactly which sections it changed and why, e.g.:
📝 Spec patched — Payments v2 Spec • Key Requirements: Added PCI-DSS compliance requirement following Alice's confirmation • Out of Scope: Removed 3DS2 after team agreed to defer it
Edit a specific section of any existing Confluence spec page directly from Slack. Claude drafts the change and shows you a preview first — you approve, revise with new instructions, or discard before anything touches Confluence.
/specbot edit Payments v2 Spec | Key Requirements | Add a requirement for idempotent payment retries
/specbot edit Payments v2 Spec | Remove all references to the legacy API
Omitting the section name lets Claude decide which section the instruction most naturally applies to.
Ask SpecBot what changed in any spec page over a given time period. It pulls the Confluence version history, compares the page before and after the period, and returns a human-readable section-by-section summary as a Slack post.
/specbot log Payments v2 Spec | last 7 days
/specbot log Payments v2 Spec | today
/specbot log Payments v2 Spec | 2026-04-01 to 2026-04-28
Supported time phrases: today, yesterday, last N days, last week, this week, or an explicit ISO date range.
Every spec-related action is recorded to a master "Spec Activity Log" Confluence page — who asked what, who ran a brainstorm, who edited a spec page directly, and when. Newest entries appear first.
Slack (slash commands / @mentions / thread messages)
↓
Flask backend (Railway)
↓ ↓
Confluence API Recall.ai API
(search + read specs) (bot joins Google Meet)
↓ ↓
Claude API (Anthropic) ← transcript chunks
↓
Confluence API
(update live draft page)
↓
Activity Logger
(append to Spec Activity Log page)
SpecBot/
├── assets/
│ └── specbot-hero.jpeg # README hero image
├── src/
│ ├── __init__.py
│ ├── app.py # Flask app — all Slack and webhook routes
│ ├── claude_client.py # Claude API — spec Q&A and proposal prompts
│ ├── confluence_client.py # Confluence REST API — search, read, write, update
│ ├── recall_client.py # Recall.ai REST API — bot lifecycle and webhook parsing
│ ├── live_proposal.py # Live session manager — buffers transcript, drives updates
│ ├── activity_logger.py # Master audit log — appends to Confluence log page
│ └── security.py # Webhook signature verification and rate limiting
├── config/
│ └── slack_manifest.yaml # Paste into Slack to create the app
├── .env.example # All required environment variables
├── .gitignore
├── CLAUDE.md
├── requirements.txt
├── railway.toml # Railway deployment config
├── SETUP.md # Step-by-step setup guide
└── README.md
| Command | Description |
|---|---|
/specbot <question> |
Answer a question from the spec |
/specbot create <title> | <brief> |
Create a new spec page from a brief description |
/specbot create <title> | <brief> | format like <page> |
Create using a specific page's format |
/specbot log <page> | <time range> |
Show what changed in a spec over a time period |
/specbot edit <page> | <instruction> |
Edit the most relevant section of a spec page |
/specbot edit <page> | <section> | <instruction> |
Edit a specific named section |
/specbot brainstorm |
Start a live Slack thread brainstorm session |
/specbot live <url> |
Join a Google Meet and write a new proposal live |
/specbot live <url> | <page title> |
Join a Google Meet and update an existing spec live |
/specbot done |
End the active brainstorm or update session |
/specbot |
Show help |
update (in brainstorm thread) |
Force an immediate proposal refresh |
done (in brainstorm thread) |
Finalise from the thread |
@SpecBot <question> |
Ask a spec question mid-conversation |
/specbot vs @SpecBot: Slash commands (/specbot) go directly to the bot — your command isn't visible in the channel, only SpecBot's response is. App mentions (@SpecBot) are normal Slack messages that tag the bot, so they're visible to everyone in the thread. Both work for spec Q&A; only slash commands support brainstorm, edit, log, and live sessions. Use @SpecBot when you're already mid-conversation in a thread and want to ask a question inline.
The Spec Activity Log Confluence page records:
| Event | Trigger | Recorded |
|---|---|---|
| ❓ Spec Question | Any /specbot or @SpecBot query |
User name, question text, specs searched |
| 🧠 Brainstorm | Session ends via done |
Who started it, mode (call/thread), link to proposal |
| ✏️ Spec Edit | /specbot edit approval, live update session ending, or direct Confluence page edit |
Editor name, page title, link |
Direct Confluence spec edits require a Confluence webhook to be configured — see SETUP.md. Edits made via SpecBot are logged automatically without a webhook.
| Component | Technology |
|---|---|
| Bot framework | Python / Flask |
| AI | Claude Sonnet (Anthropic API) |
| Live call transcription | Recall.ai |
| Spec storage | Confluence (Atlassian REST API) |
| Slack integration | slack-sdk |
| Hosting | Railway |
| Variable | Description |
|---|---|
SLACK_BOT_TOKEN |
Bot User OAuth Token from Slack app settings |
SLACK_SIGNING_SECRET |
Signing secret from Slack app Basic Information |
ANTHROPIC_API_KEY |
Claude API key from console.anthropic.com |
CONFLUENCE_BASE_URL |
e.g. https://yourcompany.atlassian.net/wiki |
CONFLUENCE_EMAIL |
Atlassian account email |
CONFLUENCE_API_TOKEN |
API token from id.atlassian.com |
CONFLUENCE_SPACE_KEY |
Key of your specs space, e.g. ENG |
RECALL_API_KEY |
API key from recall.ai dashboard |
RECALL_REGION |
Your Recall region, e.g. us-east-1 |
RECALL_WEBHOOK_URL |
https://YOUR_DOMAIN/recall/webhook |
CONFLUENCE_GOLD_STANDARD_PAGE_ID |
Optional — ID of your best spec page, used as a formatting template for proposals |
CONFLUENCE_WEBHOOK_SECRET |
Optional — shared secret to verify Confluence webhook requests |
RECALL_WEBHOOK_SECRET |
Optional — shared secret to verify Recall.ai webhook requests |
| Service | Cost |
|---|---|
| Railway | Free tier / ~£4/month Pro |
| Anthropic API | ~£0.01–0.05 per session at this scale |
| Recall.ai | ~£0.52/hour of call time |
| Slack + Confluence | Existing plan — no extra cost |
Estimated monthly total for a team under 10: under £15.
See SETUP.md for the full step-by-step guide covering API keys, Railway deployment, Slack app creation, and Confluence webhook configuration.
- Spec content is sent to the Anthropic API to generate answers. Anthropic's paid API plans include a zero data retention policy — inputs are not stored or used for training.
- Recall.ai receives meeting audio to produce transcripts. Review their data processing terms if this is a concern for your organisation.
- The bot answers spec questions strictly from Confluence content. If a spec doesn't cover a question, it says so explicitly rather than guessing.