Graphiti Memory System

Knowledge graph memory for AI agents - stores patterns, fixes, and learnings.

⚠️ REMINDER: Other Critical Skills

When this skill is invoked, also remember:

Skill	When
`quality-gate`	EVERY response with code changes — 7/7 Checklist!
`worktree`	BEFORE any code changes — Isolation check
`verification`	BEFORE claiming completion — Evidence required

GROUP ID CONVENTION

Use consistent group IDs per project:

Group ID Pattern	Usage
`project_<projectname>`	Project-specific knowledge
`global_`	Cross-project patterns

NEVER use fragmented group_ids like projectname_wip, projectname_fixes!

SESSION START PROTOCOL

Every session MUST start with context loading:

# Load project context at session start
context = mcp__graphiti__get_context(
    query="gotcha fix pattern architecture",
    group_ids=["project_<projectname>", "global_"]
)

WHEN TO SAVE MEMORY

After Fixing a Bug

mcp__graphiti__add_memory(
    name="FIX: [Brief description]",
    episode_body="Problem: [what was wrong]\nCause: [root cause]\nSolution: [how it was fixed]\nCode: [relevant snippet]",
    group_id="project_<projectname>",
    source_description="fix",
    source="text"
)

After Discovering a Gotcha

mcp__graphiti__add_memory(
    name="GOTCHA: [Brief description]",
    episode_body="Context: [when this happens]\nProblem: [what goes wrong]\nSolution: [how to avoid/fix]",
    group_id="project_<projectname>",
    source_description="gotcha",
    source="text"
)

After Learning a Pattern

mcp__graphiti__add_memory(
    name="PATTERN: [Brief description]",
    episode_body="Use case: [when to use]\nImplementation: [how to implement]\nExample: [code snippet]",
    group_id="project_<projectname>",
    source_description="pattern",
    source="text"
)

After Architecture Decision

mcp__graphiti__add_memory(
    name="ARCH: [Brief description]",
    episode_body="Decision: [what was decided]\nRationale: [why]\nAlternatives considered: [what else]\nTradeoffs: [pros/cons]",
    group_id="project_<projectname>",
    source_description="architecture",
    source="text"
)

SOURCE DESCRIPTION TAGS

Tag	When to Use
`fix`	Bug fixes with solution
`gotcha`	Known pitfalls to avoid
`pattern`	Reusable code patterns
`architecture`	Design decisions
`preference`	User/project preferences

QUERYING MEMORY

Get Formatted Context (Preferred)

mcp__graphiti__get_context(
    query="[relevant keywords]",
    group_ids=["project_<projectname>"]
)

Search Specific Nodes

mcp__graphiti__search_nodes(
    query="[search term]",
    group_ids=["project_<projectname>"],
    max_nodes=10
)

Search Facts/Relationships

mcp__graphiti__search_memory_facts(
    query="[search term]",
    group_ids=["project_<projectname>"],
    max_facts=10
)

List Recent Episodes

mcp__graphiti__get_episodes(
    group_ids=["project_<projectname>"],
    max_episodes=10
)

SUBAGENT CONTEXT LOADING

Each subagent should load relevant context at start:

Subagent Type	Query Keywords
Code Reviewer	"patterns gotchas fixes code quality"
Security	"security vulnerabilities fixes auth"
Frontend	"UI patterns hooks components fixes"
Backend	"API database patterns fixes"
Testing	"test patterns CI gotchas coverage"

WEEKLY LEARNING EXTRACTION

Every week, extract learnings from Graphiti:

"Generate weekly learning report:
  - Top 5 problems encountered
  - Patterns (>= 3 occurrences)
  - Proposed documentation updates
  - Decisions that should be recorded"

"Learn once, remember forever."

graphiti-memory