About this template
RPI / QRSP Compaction Pack
You want AI coding that stays maintainable in a brownfield repo — not a lights-off factory that ships slop until the codebase collapses.
This Claude Code microVM ships Research → Plan → Implement skills plus intentional compaction, so context stays in the smart zone and humans review high-leverage artifacts before code.
Undeniable demo
- Fork and log into Claude Code with your key
- Run research on a real bug or small feature
- Approve the research → produce a vertical-slice plan
- Implement phase 1 only; compact status back into the plan
- Do not merge — open a draft PR if you want
What's installed
| Path | What |
|---|---|
~/.claude/skills/rpi-research/ | Compact codebase research |
~/.claude/skills/rpi-plan/ | Implementation plan + verify steps |
~/.claude/skills/rpi-implement/ | Phase-by-phase execute |
~/.claude/skills/intentional-compaction/ | Fresh-window progress distill |
~/vault/rpi.md | Human-gate defaults |
BYOK
ANTHROPIC_API_KEYor Claude Code login- Optional
GH_TOKENfor PR context
Who it's for
Staff/principal engineers running agents in complex codebases who refuse lights-off.
Not for
Unattended merge bots. Ocean-boiling "just token harder" workflows.
Pricing
jurniti microVM from $25/mo. 30-day money-back on first purchase. No free trial.
Positioning
Anti-lights-off Software Factory: agents draft; humans still read and merge. Aligned with advanced context engineering / program design discipline.