DSOM Automated State Sync
1. Vectorized Memory Tiering
Instead of pushing an entire project history into every prompt, DSOM categorises information into distinct operational layers:
* Active Layer: Houses only the immediate, high-priority variables for the current task.
* Compressed Layer: Stores foundational rules and past decisions as dense, key-value indices (.agents/brain/current_state.dsom).
* Archival Layer: Offloads long-term data to the Git history and Sovereign Markdown Palace.
2. Semantic Compaction (Token Distillation)
DSOM applies a "distillation loop" using an automated GitHub Action.
When a Pull Request is merged, .github/workflows/dsom-pr-sync.yml triggers .github/scripts/action_update_dsom.py.
This script calls an LLM (e.g. OpenAI) to review the PR diff and update .agents/brain/current_state.dsom strictly in OKF v0.1 format, eliminating redundant conversational fluff and appending only critical architectural decisions.
3. Configuration
- Ensure the
OPENAI_API_KEYsecret is available to the GitHub Action. - The state is persisted in
.agents/brain/current_state.dsom.
Deep State of Mind (DSOM) For My AI Protocol | Harisfazillah Jamel (LinuxMalaysia) | 2026-08-14 Standard: UK English | DBP-standard Bahasa Melayu Malaysia (Piawai) | GNU General Public License v3.0