From ctx
This skill should be used when the user asks to "clean up memory", "sync ctx and memory", "review memory", "deduplicate nodes", "fix memory conflicts", "ctx cleanup", "memory maintenance", "reconcile memory systems", or mentions stale nodes, memory bloat, or keeping ctx and MEMORY.md in sync.
npx claudepluginhub zate/cc-plugins --plugin ctxThis skill is limited to using the following tools:
Reconcile Claude Code's built-in MEMORY.md system with the ctx knowledge graph. Detect duplicates, stale nodes, conflicts, and bloat across both systems, then resolve them with user approval.
Organizes, extracts, prunes, and verifies Claude Code persistent memory files: MEMORY.md index, topic extraction, staleness detection, accuracy checks. Use near 200-line limit, after insights, or project changes.
Optimizes Claude Code memory files in 4 interactive steps: removes duplicates by cross-referencing CLAUDE.md/rules, migrates entries to persistent configs, compresses, and validates with cleanup. Run in main conversation to declutter memory.
Optimizes Claude Code auto-memory files in 4 interactive steps: removes duplicates, migrates rules to CLAUDE.md/rules, compresses entries, validates cleanup. Reduces token count 30-50%.
Share bugs, ideas, or general feedback.
Reconcile Claude Code's built-in MEMORY.md system with the ctx knowledge graph. Detect duplicates, stale nodes, conflicts, and bloat across both systems, then resolve them with user approval.
$1 — Project name (optional, auto-detected from git)--global — Include all projects, not just the current oneThis is a phased command — run each phase in order, presenting findings to the user between phases.
Gather the complete state of both memory systems.
Actions:
basename $(git rev-parse --show-toplevel))sh "${CLAUDE_SKILL_DIR}/scripts/analyze-sync.sh" "$PROJECT"
~/.claude/projects/ (search for directory matching project name)ctx list --tag "project:$PROJECT" --format json
project:global):
ctx list --tag "project:global" --format json
If --global was passed, also inventory all other projects:
~/.claude/projects/*/memory/project:* tags in ctx via ctx tagsSave the inventory mentally — it feeds all subsequent phases.
Analyze the inventory for seven categories of issues. For each issue found, record: the issue type, affected items (node IDs / MEMORY.md line numbers), severity (high/medium/low), and recommended action.
Issue categories (consult references/sync-strategies.md for detailed detection patterns):
Cross-system duplicates — Same knowledge in both MEMORY.md and ctx
Intra-ctx duplicates — Two+ ctx nodes with near-identical content
Stale working nodes — tier:working nodes from completed tasks
Pinned tier bloat — Too many pinned nodes consuming context budget
Conflicts — Contradictory information across systems
Orphaned content — Knowledge for projects that no longer exist or are inactive
Missing structure — Nodes without tier tags, without project tags, or with empty content
Present findings to the user in a structured report. Group by severity (high first).
Report format:
## Memory Sync Report — [project]
### Token Budget
- MEMORY.md: ~X tokens (Y lines)
- ctx pinned: X tokens (Y nodes)
- ctx working: X tokens (Y nodes)
- Total auto-loaded: X tokens
### Issues Found
#### HIGH: Stale Working Nodes (N found)
| Node ID | Content Preview | Created | Recommendation |
|---------|----------------|---------|----------------|
| 01KK... | Sprint complete... | Mar 15 | -> tier:reference |
#### MEDIUM: Cross-System Duplicates (N found)
| MEMORY.md Line | ctx Node | Content | Recommendation |
|---------------|----------|---------|----------------|
| L5-7 | 01KH... | Release rules | Remove from MEMORY.md |
... (all categories)
### Summary
- X issues found (Y high, Z medium, W low)
- Estimated token savings: ~N tokens
MANDATORY: After presenting the report, MUST use the AskUserQuestion tool (not plain text) to ask how to proceed. Use this exact configuration:
Do NOT proceed to Phase 4 without this user response. Do NOT present these options as text — use the AskUserQuestion tool.
Execute approved fixes. For each fix, use the appropriate mechanism:
ctx node actions:
ctx tag <id> tier:reference && ctx untag <id> tier:workingctx tag <id> tier:off-context && ctx untag <id> tier:workingctx delete <id> (only for true duplicates or empty nodes)<ctx:supersede old="ID1" new="ID2"/> when one node replaces anotherctx tag <id> project:name or ctx tag <id> tier:referenceMEMORY.md actions:
Cross-system actions:
After each batch of fixes, show a summary of what changed.
After remediation:
ctx status to show the new stateAfter cleanup, advise the user on keeping systems in sync:
The using-ctx skill already handles this — it instructs the agent to cross-check before writing to either system. If the user feels drift is happening, the skill instructions may need strengthening.
Periodic cleanup — Run /ctx:cleanup quarterly or when starting major new work.
Hook option — For persistent drift, a PostToolUse hook on MEMORY.md writes can enforce cross-checking. See references/sync-strategies.md for the hook design. Only recommend this if the user reports ongoing issues after multiple cleanups.
references/sync-strategies.md — Detailed sync strategies, division of labor rules, issue detection patterns, token budget guidelines, and hook-based sync designscripts/analyze-sync.sh — Dumps both memory systems for comparison (run with project name argument)