RaiSE v2.2.2

RaiSE is a lean methodology and toolkit for AI-assisted software engineering. If you’re new here — welcome. This is a good release to start with.

v2.2.2 focuses on reliability. No new features. Just making the ones you already use work better, especially when you change machines, upgrade, or work across multiple projects.

Seamless upgrades

Previously, after upgrading RaiSE (pipx upgrade raise-cli), your project’s skills stayed at the old version until you manually ran rai init. Easy to forget, hard to diagnose.

Now rai session start detects the mismatch and syncs automatically. If you’ve customized a skill, your version is preserved. If you haven’t, it updates silently.

$ rai session start --project . --context
Skills synced to 2.2.2 (3 updated, 1 new)

No manual steps. Your tools stay current.

Session continuity

Sessions are how RaiSE maintains context between working periods — what you were doing, what patterns you’ve learned, what decisions were made. Three bugs were affecting this continuity:

Counter resilience: Changing machines or cloning fresh could reset your session counter, breaking references to previous sessions. The counter now has a filesystem fallback and never resets silently.
Migration accuracy: Moving to the new per-session storage format could create a phantom session entry, generating false warnings on every start. Fixed — migrations now preserve the correct session identity.
Pattern tracking: Patterns captured during session close were recorded with placeholder IDs instead of their real identifiers. Now tracks accurately.

These were silent issues. You might not have noticed them, but they affected the quality of your session history.

Graph integrity

The knowledge graph is the backbone of RaiSE context — it connects governance documents, patterns, epics, stories, and code structure. If two source files generated the same node ID, the graph builder would silently discard one.

Now it stops and tells you exactly where the collision is. We believe in Jidoka — stop on defects, don’t let them propagate.

Quality of life

CLI quick reference now documents all journal entry types (decision, insight, task_done, note), so your AI agent doesn’t have to guess.
MCP server health checks report “not connected” instead of showing raw error traces.
Internal skill documentation corrected for pattern type references.

Source code is now public

The GitHub repository is open.

We believe a framework that asks you to trust AI-assisted development should itself be fully transparent. Read the code, file issues, judge the craft.

Every skill, every graph builder, every session tracker — it’s all there.

New to RaiSE? pipx install raise-cli && rai init

Already using it? pipx upgrade raise-cli

Full changelog: CHANGELOG.md

Found something? Open an issue