make-plan: Memory-Grounded Planning
A good plan starts from the current state, not a blank slate. Memory tells you what's been decided, what's blocking progress, and what was left mid-session.
Planning Workflow
1. Load context from memory
# What was done recently?
python3 {SKILL_DIR}/scripts/memory_store.py timeline --hours 168 --limit 30
# What's in progress / blocked?
COWORK_MEM_DB=~/mnt/.claude/.cowork-mem/memory.db \
python3 {SKILL_DIR}/scripts/vector_search.py "in progress blocked next step" --limit 10
# What did the last session summary say?
python3 {SKILL_DIR}/scripts/memory_store.py search "session summary" --type summary --limit 5
2. Check for known errors or constraints
python3 {SKILL_DIR}/scripts/memory_store.py search "error bug blocked" --type error --limit 5
3. Build the plan
Structure the plan as:
- Context: What we know from memory (1-3 bullets, specific)
- Goal: What we're trying to accomplish
- Steps: Numbered, specific, actionable — reference actual files and systems
- Blockers: Known issues from memory that affect the plan
- Open questions: Things we'd need to investigate
4. Save the plan as a decision
python3 {SKILL_DIR}/scripts/memory_store.py add decision \
"Plan: <summary of steps>" --tags "plan,session-goal"
Planning Tips
- Plans built from memory are more accurate than plans from scratch — use it
- Surface blockers from memory explicitly; don't let them be surprises mid-sprint
- If memory has conflicting decisions, flag it for the user to resolve
- Short plans (3-5 steps) are better than exhaustive ones — you can always plan the next thing after