We built Agent Session Kit (ASK) while experimenting with AI-assisted development workflows.
One thing we noticed is that AI coding sessions tend to be fast and bursty, which increases the chances of workflow drift.
Examples we ran into repeatedly:
• commits landing in the wrong branch • session context docs drifting out of date • resuming sessions with stale assumptions • inconsistent environments between machines
ASK adds lightweight guardrails around normal Git workflows using hooks, validation scripts, and smoke tests.
The goal is to make good workflow habits automatic rather than relying on memory.
Curious to hear how others are managing workflow discipline when coding with AI agents.
qarau•2h ago
The issue wasn’t the code generation itself. It was workflow discipline around the code.
AI-assisted sessions move fast and often run long. Over time we noticed recurring problems like:
- commits landing in the wrong branch
- session context docs drifting out of date
- resuming sessions with stale assumptions
- environment inconsistencies between machines
- forgetting validation steps before pushing
None of these are new problems, but AI makes them happen more frequently because iteration speed is so high.
We didn’t want to introduce another heavy dev platform, so we built a small toolkit that adds lightweight guardrails around normal Git workflows.
That toolkit became Agent Session Kit (ASK).
ASK installs a set of helpers around a repo including:
- Git hook templates - session validation scripts - environment smoke tests - session documentation helpers - workflow checks before commit/push
The goal is simple:
Turn workflow discipline into default behavior rather than relying on developer memory.
If something is off (wrong branch, missing session updates, failing checks), the workflow catches it before the change lands in the repo.
We originally built ASK as internal tooling while experimenting with AI-assisted development workflows and decided to open-source it so other teams don’t have to rebuild the same guardrails.
The project is intentionally simple:
- no servers - no dashboards - just scripts and Git integration
If you’re experimenting with AI coding agents or long interactive dev sessions, this might be useful.
Feedback and suggestions welcome.