FlowSentinel

## Inspiration

We observed that even with documented branching strategies, human error remains the #1 cause of environment desynchronization. We wanted to see if agentic AI could replace manual "gatekeeping" and rigid, brittle Bash scripts.

## What it does

FlowSentinel uses a single YAML configuration to define the "Laws of the Repo." A GitLab Duo Agent monitors every push and merge request, using LLM reasoning to interpret complex promotion conditions (like "is this branch aligned with PPD?") that are usually difficult to script.

## How we built it

We utilized the GitLab Duo Agent **. We configured a Custom Agent with access to the repository context and defined an **Agentic Flow that triggers on MR events. The agent uses the ai-catalog to reference our specific governance rules.

## Challenges we ran into

The biggest hurdle was "Regex Escaping" within the YAML to ensure the AI interpreted Git patterns literally. We also had to fine-tune the Agent's "tool use" so it wouldn't just guess branch alignment but would actually execute a git merge-base check via the API.

## Accomplishments that we're proud of

We successfully built a system where no code was written to check the code. The compliance logic lives in a readable YAML file, and the "engine" is a managed AI agent that provides conversational feedback to developers.

## What we learned

AI Agents are significantly better at "soft governance" than hard-coded scripts. When a developer makes a mistake, the Agent doesn't just fail a job; it explains why based on the YAML rules, acting as a coach rather than a barrier.

## What's next for FlowSentinel

Integration with GitLab Duo Root Cause Analysis to automatically suggest the exact git merge commands needed to bring a "non-aligned" branch back into compliance.

Built With

• ansible
• gitlab
• python
• unix