Strategies for Automating Issue & PR Labeling: Rule-based vs. AI-Native

[eazyhozy]

Purpose

This discussion explores options for automating issue and pull request management for this project (related to #11).

As the project becomes open-source, we want a simple and scalable way to handle contributions, starting with issue and PR labeling.

---

Approaches Seen in Other Projects

• Rule-based automation (e.g. actions/labeler)

  • Common in large projects
  • Reliable and easy to maintain

• AI-powered automation (e.g. dosubot)

  • Provides context-aware labeling and PR feedback
  • More powerful, but often requires additional setup or paid plans

---

Proposed Direction (Open for Discussion)

A possible phased approach:

1. Short-term: Use a lightweight rule-based setup (actions/labeler) for basic structure.
2. Long-term: Evaluate AI-based tools once the project and contribution volume grow.

This is a proposal, not a decision.

---

Feedback Requested

• Are there AI automation tools (open-source or SaaS) you recommend?
• Which tools work well for:
  • Context-aware labeling
  • Automated PR comments or reviews
  • Issue / PR triage

Please share any experiences or suggestions.

[em3s]

From what I’ve seen, Dosu seems to be working well for SlateDB.
I support a phased approach - starting with what’s practical today and gradually moving toward more AI-native automation.

@eazyhozy

[em3s]

While working on the CI setup, we also discussed strategies for automating Issue & PR labeling.

Related context: #11 (comment)

We initially considered a rule-based approach, but Dosu was integrated quickly (thanks to @eazyhozy), allowing us to experiment directly with an AI-powered workflow. (This is still an early-stage choice and not a fixed decision.)

[eazyhozy]

FYI: How dosu is connected to GitHub

At the moment, the dosu instance connected to this repository is authorized via the personal GitHub account @eazyhozy.
This discussion will be updated if the connection setup changes.

[em3s]

Closing—currently using Dosu-based labeling, which is working well. Further improvements can be discussed separately.