ADR-0010: Research Data Sync

July 31, 2026 · View on GitHub

Status

accepted

Date

2026-03-25

Context

Runtime data (knowledge.json, identity.md, history/, etc.) is stored in MOLTBOOK_HOME but is not under git version control. We want to version-control it for research purposes, but mixing it into the main repository causes the following problems:

  • Episode logs (logs/*.jsonl) are a prompt injection vector (ADR-0007)
  • Runtime data and source code commit histories become interleaved
  • Risk of accidentally committing sensitive files like credentials.json

Decision

Rsync only safe runtime data to ~/MyAI_Lab/contemplative-agent-data/ (a separate repository), then automatically git commit + push after distill execution.

Sync Targets

  • knowledge.json, identity.md, agents.json
  • history/identity/*, history/knowledge/*
  • skills/*, rules/*, meditation/results.json
  • reports/comment-reports/* (from the project repository)

Exclusions

  • logs/*.jsonl — Prompt injection vector (ADR-0007)
  • credentials.json — API keys
  • rate_state.json, commented_cache.json — Ephemeral data with no research value

Sync Timing

Runs as a post-processing step of the distill command. No additional launchd plist required. Manual execution: contemplative-agent sync-data.

Consequences

  • Runtime data evolution can be tracked via git log
  • The main repository stays focused on source code only
  • Risk of accidentally publishing episode logs is eliminated