Retrieve and utilize stored memories for AI employees in Slack environments. Efficiently searches and loads relevant context (channels, users, projects, decisions, meetings) from organized memory storage to inform responses. Use this when answering questions that require historical context, user preferences, project status, or any previously stored information. Works with slack-memory-store storage system.
This skill enables AI employees to efficiently retrieve and utilize stored memories to provide context-aware responses in Slack conversations.
Retrieve relevant memories from {memories_path} to inform the next response with appropriate context about people, projects, decisions, preferences, and work history.
Every memory retrieval follows this pattern:
User (in #마케팅팀): "Q4 전략 어떻게 되고 있어?"
Step 1: Context Analysis
- Channel: #마케팅팀 (C123)
- Keywords: Q4, 전략
Step 2: Read Index
view {memories_path}/index.md
→ See Recent Updates, locate Q4-related items
Step 3: Load Relevant Files
view {memories_path}/channels/C123_마케팅팀.md
view {memories_path}/projects/Q4전략.md
view {memories_path}/meetings/ (find Q4 meetings)
Step 4: Respond
Synthesize information from channel context, project status, and meeting notes
The memory system uses a hybrid approach:
channels/C123_마케팅팀.md - Channel guidelines, preferences, static infousers/U456_김철수.md - User profile, communication styleprojects/신제품런칭.md - Project discussionsdecisions/AWS전환_20251117.md - Important decisionsmeetings/2025-11-17-Q4회의.md - Meeting notesmisc/마케팅팀_일상_20251117.md - Casual conversations{memories_path}/
├── index.md # START HERE - navigation and stats
├── channels/ # Channel profile files (one per channel)
├── users/ # User profile files (one per user)
├── projects/ # Project topic files (multiple)
├── tasks/ # Task records
├── decisions/ # Decision records (date-stamped)
├── meetings/ # Meeting notes (date-stamped)
├── feedback/ # User feedback
├── announcements/ # Important announcements
├── resources/ # Internal docs and guides
├── external/news/ # External information
└── misc/ # Uncategorized conversations
CRITICAL: Every retrieval session must begin by reading index.md:
view {memories_path}/index.md
The index provides:
This one-time read gives you the complete map of available memories.
Extract context from the conversation:
Channel Context:
Message in #마케팅팀
→ Load: {memories_path}/channels/C123_마케팅팀.md
→ Check related_to metadata for connected info
User Context:
DM from @chulsoo
→ Load: {memories_path}/users/U123_김철수.md
→ Get communication_style, preferences
Project Context:
Question about "신제품 런칭"
→ Load: {memories_path}/projects/신제품런칭.md
→ Check milestones, status, participants
Keyword Context:
Question mentions "결정", "승인"
→ Search: {memories_path}/decisions/
→ Find relevant decision files
Tier 1: Always Load (if relevant)
Tier 2: Load as Needed
Tier 3: Load Selectively
Don't over-fetch. Use directory listings first:
view {memories_path}/projects/
# See available projects before loading specific files
When responding in a channel:
# 1. Load channel context
view {memories_path}/channels/{channel_id}_{channel_name}.md
# 2. Check for channel guidelines
# Extract: tone, response_time, key_topics
# 3. Apply guidelines to response
# Adjust tone, format based on channel preferences
When responding to a specific user:
# 1. Load user profile
view {memories_path}/users/{user_id}_{name}.md
# 2. Check communication_style
# Extract: tone, detail_level, preferences
# 3. Personalize response
# Match user's preferred style and detail level
When asked about project status:
# 1. Find project file
view {memories_path}/projects/
view {memories_path}/projects/{project_name}.md
# 2. Check metadata
# status, priority, milestones, participants
# 3. Get related info
# Check related_to for decisions, meetings
# 4. Provide comprehensive update
# Current status + recent activity + next steps
When asked about past decisions:
# 1. Search decisions
view {memories_path}/decisions/
# 2. Load relevant decision file
view {memories_path}/decisions/{decision_name}.md
# 3. Extract key info
# decision_makers, rationale, alternatives_considered
# 4. Explain context
# Why decision was made + alternatives + outcome
When asked about completed work:
# 1. Check completed tasks
view {memories_path}/tasks/completed/
# 2. Filter by assignee/date
# Look for relevant assignee, date range
# 3. Summarize work
# List tasks + effort + outcomes
Follow the trail of related information:
# In project file:
---
related_to:
- decisions/기술스택선택.md
- meetings/2025-10-20-기획회의.md
---
Load related files to build complete context.
Use metadata to filter without reading entire files:
# List directory first
view {memories_path}/projects/
# Check filenames and metadata
# Only load files matching criteria:
# - status: in_progress
# - priority: high
# - participants: includes current_user
Consider time-sensitivity:
# Recent Updates in index.md
→ Shows 10 most recent changes
→ Focus on these for "latest" questions
# File metadata: created, updated
→ Check dates to prioritize fresh info
Use tags for discovery: