MindStone for Pi
MindStone for Pi
Section titled “MindStone for Pi”MindStone for Pi (MS4PI) is the Pi substrate edition of MindStone Agent continuity.
It gives a Pi coding-agent instance continuity across sessions: identity, user context, memory, checkpoint discipline, semantic recall, compaction handoff, role adoption, and optional Synapse communication.
The design principle is that memory is not a feature. It is a layered continuity system. LOG.md is audit, vectors are recall, .handoff.md is a rich continuity prompt for compaction survival, identity is orientation, and structured memories are durable facts. None is sufficient alone; together, they let the agent resume with better judgment.
In MS4PI, a checkpoint is a governed synthesis step that promotes session experience into durable LOG, memory, transcript, and index state. A handoff is a detailed prompt for the next agent state, richer and more operational than a generic context summary.
What MS4PI adds
Section titled “What MS4PI adds”| Capability | Stock Pi | With MS4PI |
|---|---|---|
| Identity | Session-local context | First-person IDENTITY.md and USER.md loaded into turns |
| Memory | Session history exists, but not as curated memory | Markdown memory files, MEMORY.md, LOG checkpoints, and archived transcripts |
| Recall | Mostly current context unless manually supplied | Prompt-time recall from memory and archived Pi session JSONL |
| Checkpointing | Manual summaries if any | MS4-style governed synthesis: LOG + memory proposal + archive/embed verification |
| Compaction | Summary-oriented context reduction | Rich .handoff.md prompt and recent-tail continuity mechanics |
| Role adoption | Prompt-based only | /act-as <role> and /end-role role-span discipline |
| Agent communication | No built-in family channel | Optional Synapse commands/tools/digest support |
Current status
Section titled “Current status”MS4PI is early but functional. The public repo includes:
- Pi package manifest and install/update commands
- identity/user/log/memory context injection
- role adoption
- checkpoint, handoff, and end-session commands
- Python recall/indexing stack
- semantic recall over memory and archived Pi transcripts
- compaction handoff scaffolding
- optional Synapse integration
Do not claim full MS4CC parity until fresh install, command reload, compaction, and Synapse paths are validated on the target machine.