SlackSync

Inspiration

Engineering teams already coordinate in Slack, but the actual work - reviewing a PR, writing tests, updating docs, figuring out why CI went red - happens in a dozen other tabs. Every context switch is lost time and lost context. We wanted the place where the team already talks to also be the place where the work gets done and audited.

What it does

SlackSync routes /sync commands and App Home actions through a deterministic Maestro router into four specialized agents:

Reviewer analyzes PR risk and posts severity-ranked review notes with a merge-readiness score.
Tester drafts syntax-valid test files for a target source file and estimates coverage.
Scribe prepares README or changelog updates from repository activity.
Watchdog triages CI failures, names the likely root cause, and suggests the fix path.

Every result is a typed data contract that a dedicated Block Kit formatter layer renders into a threaded Slack message, and every result carries the MCP and Real-Time Search context it was based on, so the answer is auditable.

How we built it

Backend: FastAPI with a Maestro command router, four agent modules, typed Pydantic schemas, a Slack signature-verification layer with a five-minute replay window, and a Block Kit formatter that keeps presentation out of agent logic.
Integration boundaries: an MCP context adapter for GitHub/Jira and a Real-Time Search adapter for Slack channel context. In demo mode they return deterministic, clearly labeled fixtures so the system runs with zero secrets; in real mode they are where live clients plug in.
Frontend: a React + Vite + TypeScript judge console that mirrors the Slack experience, calls the FastAPI route when available, and otherwise falls back to deterministic data so it can be hosted statically on GitHub Pages.
Shared contracts: a TypeScript contracts package keeps the UI and API payloads aligned.
Quality: GitHub Actions CI runs backend pytest, frontend build/test, and a secret scan on every push.

Required technologies

Slack Agent / Agent Builder: an installed Slack app answers the /sync slash command through the FastAPI webhook and returns Block Kit.
MCP: services/api/app/integrations/mcp_context.py attaches GitHub/Jira MCP evidence to routed results, asserted by tests.
Real-Time Search: services/api/app/integrations/rts_search.py attaches matching Slack channel context, asserted by tests.

Challenges we ran into

Keeping the project demo-able with zero secrets while still proving the real Slack path was the central tension. We solved it by making agent outputs typed data contracts and treating Slack signature verification, MCP, and RTS as explicit, testable boundaries - real where it counts, deterministic where a secret would otherwise be required, and clearly labeled either way.

Accomplishments we're proud of

A working /sync command live in a Slack sandbox, four agents behind one deterministic router, a Block Kit formatter layer, MCP and RTS evidence on every answer, a public judge console, green CI, and an end-to-end demo that needs no secrets to run.

What we learned

Designing agent outputs as data contracts first - and pushing all Slack formatting into one layer - made the system far easier to test, demo, and extend than wiring Block Kit through the agents would have.

What's next

Live MCP and Real-Time Search clients behind the existing adapters, App Home dashboards and interactive approvals, and a path toward a Slack Marketplace listing once the product is hardened.

Track

Primary: New Slack Agent. The implemented MCP and Real-Time Search evidence adapters also align with those category prizes. Slack Agent for Organizations remains a stretch and is not claimed until a real Marketplace submission exists.

Impact

SlackSync reduces context switching across Slack, GitHub, CI, Jira, and docs by turning Slack into the command surface and the audit trail for agentic engineering workflows.