Stats
Actions
Tags
From memoryrelay
Use when starting a new conversation or task that needs persistent memory, storing or retrieving information across sessions, or working within a project that uses MemoryRelay.
How this skill is triggered — by the user, by Claude, or both
Slash command
/memoryrelay:memory-workflowThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Follow this order every time. Skipping steps causes orphaned memories.
Follow this order every time. Skipping steps causes orphaned memories.
| Step | Call | Purpose |
|---|---|---|
| 1 | project_context(project) | Load hot-tier memories, active decisions, adopted patterns |
| 2 | session_start(title, project) | Begin tracking work (returns session_id) |
| 3 | decision_check(query, project) | Check existing decisions before architectural choices (see decision-tracking skill) |
| 4 | pattern_search(query) | Find established conventions (see pattern-management skill) |
| Action | Tool | Notes |
|---|---|---|
| Save info | memory_store(content, metadata) | Always set deduplicate=true |
| Search | memory_recall(query, limit?, threshold?) | Semantic search across memories |
| Delete | memory_forget(id_or_query) | By ID or fuzzy search |
| Browse | memory_list(limit, offset) | Chronological listing |
| Read one | memory_get(id) | Fetch by exact ID |
| Edit | memory_update(id, content) | Correct or expand existing |
| Bulk save | memory_batch_store(memories[]) | Efficient for multiple items |
| Build prompt | memory_context(query, token_budget) | Token-aware context window (see entity-and-context skill) |
| Upgrade | memory_promote(id, importance, tier) | Move temporary to long-term |
For architectural choices, use decision_record instead of memory_store — see the decision-tracking skill.
For reusable conventions, use pattern_create instead of memory_store — see the pattern-management skill.
Call session_end(session_id, summary) with a meaningful summary. This becomes the historical record.
Always pass deduplicate=true on memory_store and memory_batch_store. The default threshold is 0.9 similarity. Skipping this clutters search results with near-duplicates.
Always include category and tags in metadata:
metadata: { "category": "technical", "tags": "auth, api", "source": "code-review" }
Categories: technical, preference, credential, decision. Consistent metadata makes filtering reliable.
| Tier | Retention | Use for |
|---|---|---|
hot | Always loaded by project_context | Critical project facts |
warm | Retrieved by search | General knowledge |
cold | Archived, low priority | Historical notes |
Use memory_promote(id, importance, tier) to upgrade a memory. Set importance near 1.0 for critical items.
| Mistake | Fix |
|---|---|
| Storing without a session | Always call session_start first |
Skipping deduplicate=true | Set it on every memory_store call |
Using memory_store for decisions | Use decision_record instead (see decision-tracking skill) |
Using memory_store for conventions | Use pattern_create instead (see pattern-management skill) |
| No category/tags in metadata | Always include both for searchability |
| Storing API keys or passwords | Blocklist auto-rejects these; use a secrets manager |
npx claudepluginhub memoryrelay/memory-relay-skills --plugin memoryrelayManages persistent semantic memory across sessions: store/retrieve knowledge/TODOs/issues, hybrid semantic search, hierarchy/tags organization, and maintenance tools.
Manages AI memories: saves decisions/patterns/facts with categories/tags, searches context, lists/deletes entries, rescans project architecture via MCP tools and /remember /forget.
Proactively saves decisions, conventions, bugs, discoveries, and preferences to persistent Engram memory across sessions using mem_save and related tools.