Memory Log

Build or refresh durable project memory across sessions.

The memory-log workflow creates a concise durable record of decisions, assumptions, open questions, and next-session priorities.

How it works

The workflow consolidates long-running context in a structured format:

  1. Artifact review — reads recent outputs, plans, and session notes.
  2. Signal extraction — captures durable items (decisions, assumptions, blockers).
  3. Status update — marks assumptions as validated, untested, or rejected.
  4. Resume prep — writes a next-session bootstrap list.

This prevents context loss across days, handoffs, or interrupted runs.

Real examples

Example 1 (project continuity):

/memory-log "llm-evals-roadmap"
bohr memory-log "llm-evals-roadmap"

Example 2 (after a major research sprint):

/memory-log "enterprise-rag-reliability"
bohr memory-log "enterprise-rag-reliability"

Expected output

Primary output file:

  • notes/<slug>-memory.md

Typical structure:

  • durable decisions and rationale
  • active assumptions with status
  • open questions and blockers
  • next-session top priorities

Example snippet:

## Decisions
- [D3] Standardize on evaluation set v2 for all latency comparisons.
  - Why: v1 showed leakage risk in two categories.

## Next Session Bootstrap
- Re-run contradiction pass on executive summary claims.
- Validate assumption A4 with experiment workflow.

Customization

You can tailor memory logs by specifying:

  • focus area — “track only model-choice decisions”
  • granularity — concise summary vs detailed ledger
  • time scope — “include only last 2 weeks”
  • handoff mode — include checklist optimized for another teammate

When to use

Use this workflow for long-running projects where context continuity matters across days, teammates, and handoffs.

For single one-off tasks, this is usually unnecessary.