Six-Layer Memory

Use this skill when the user wants a workspace to keep memory proactively instead of relying on chat context alone.

What this skill sets up

• HOT: memory/SESSION-STATE.md
• WARM: local vector/index refresh when source material changes
• COLD: memory/decisions/
• CURATED: MEMORY.md plus daily logs
• LOCAL-AUTO: optional local maintenance log and digest checks, with no external API calls

Workflow

1. Pick the target workspace.
2. Run scripts/install_workspace.sh <workspace>.
3. Review the created files before adding any schedule.
4. Add periodic execution only after the user confirms the schedule:

0 6 * * * /usr/bin/python3 <workspace>/memory/auto_memory_6layer.py --workspace <workspace> --daily --source memory-daily

5. Validate with memory/check_memory_layers.sh.

Bundled scripts

• scripts/install_workspace.sh
• scripts/auto_memory_6layer.py
• scripts/memory_writer.py
• scripts/check_memory_layers.sh

Notes

• This skill is designed per workspace. Repeat installation for each agent workspace.
• Do not overwrite HOT state with synthetic “all good” status text.
• Prefer one canonical HOT file path: memory/SESSION-STATE.md.
• Keep MEMORY.md for durable facts only. Put operational notes elsewhere.
• Do not store API keys, access tokens, cookies, login state, or customer private data in public skill files, shared templates, or examples.
• If a user wants cloud sync, treat it as a separate private extension: explain the risks, ask for explicit confirmation, and keep keys outside the published skill.

References

• Local-first guide: references/local-first-guide.md