Agent Memory - Knowledge Sharing

WHY Use Memory

Memory persists across ALL agent sessions. What you learn today helps other agents (and future you) tomorrow.

• Avoid repeating mistakes
• Reuse solutions that worked
• Maintain context across sessions
• Share knowledge between agents

WHEN to Search Memory

Before Starting Work

# Search for prior work on this ticket/topic
search_memory(query: "ticket #114", limit: 5)
search_memory(query: "daisyUI theme configuration", limit: 5)

When Stuck or Blocked

# Search for similar problems or solutions
search_memory(query: "Stacked PRs phase dependency", limit: 5)
search_memory(query: "rails credentials error", limit: 5)

Before Making Decisions

# Search for prior decisions on this topic
search_memory(query: "authentication", memory_type: "decision", limit: 5)
search_memory(query: "database schema", memory_type: "decision", limit: 5)

When Encountering Errors

# Search for previous solutions to this error
search_memory(query: "ActiveRecord not found error", memory_type: "error", limit: 5)

WHEN to Store Memory

After Learning Something New

store_memory(
  content: "Learned that daisyUI themes are configured in application.tailwind.css with @plugin directive",
  memory_type: "fact",
  ticket_id: 89
)

After Fixing a Bug

store_memory(
  content: "Fixed: Tailwind v4 requires @import 'tailwindcss' instead of @tailwind directives. Error: 'Unknown at rule @tailwind'",
  memory_type: "error",
  ticket_id: 89,
  metadata: { error_message: "Unknown at rule @tailwind", solution: "Use @import 'tailwindcss'" }
)

After Making a Decision

store_memory(
  content: "Chose daisyUI winter theme as default over emerald because better color contrast for accessibility",
  memory_type: "decision",
  ticket_id: 89
)

After Implementing a Pattern

store_memory(
  content: "When fixing existing PRs, always checkout the PR's branch using 'gh pr view X --json headRefName' instead of creating new branch",
  memory_type: "instruction",
  ticket_id: 98
)

After Completing Work

store_memory(
  content: "Implemented daisyUI theme polish with 3 themes: winter (default), dracula (dark), corporate. All AVO cards converted to daisyUI components (card, badge, progress, alert). Theme switcher component created with Stimulus controller.",
  memory_type: "summary",
  ticket_id: 89
)

Memory Types

Quick Reference

Best Practices

BE SPECIFIC

# Good
store_memory(content: "daisyUI 5.5.14 requires Tailwind CSS 4.1.18+ for @plugin syntax", memory_type: "fact")

# Too vague
store_memory(content: "Check daisyUI version compatibility", memory_type: "fact")

INCLUDE CONTEXT

store_memory(
  content: "For AVO cards, use daisyUI 'card' component with 'bg-base-100' for theme-aware backgrounds",
  memory_type: "instruction",
  ticket_id: 89,
  metadata: { component: "avo_card", theme: "daisyui" }
)

LINK TO TICKETS

store_memory(
  content: "git-workflow skill supports stacked PRs for phased development (Phase 1 → Phase 2 → Phase 3)",
  memory_type: "fact",
  ticket_id: 97
)

Examples by Memory Type

decision - Architectural Choices

# After choosing a library
store_memory(
  content: "Chose bun over npm/yarn because 10x faster install times and native ESM support",
  memory_type: "decision",
  ticket_id: 81
)

# After choosing an approach
store_memory(
  content: "Decided to use skills pattern (from ticket #98) instead of inline prompts for persistent workflows",
  memory_type: "decision",
  ticket_id: 114
)

error - Bugs and Solutions

# After fixing a specific error
store_memory(
  content: "Fixed: 'ActionController::UnknownFormat' - Add 'format.js { render json: {...} }' to respond_to block",
  memory_type: "error",
  ticket_id: 105,
  metadata: { error_class: "ActionController::UnknownFormat", solution: "Add format.js response" }
)

# After resolving configuration issues
store_memory(
  content: "Fixed: daisyUI components not rendering - Needed to run 'bun run build:css' to compile Tailwind v4",
  memory_type: "error",
  ticket_id: 89
)

instruction - Reusable Patterns

# Workflow guidance
store_memory(
  content: "When implementing epic subtasks, always check for auto-block dependencies via ticket.dependencies before starting work",
  memory_type: "instruction",
  ticket_id: 96
)

# Code pattern
store_memory(
  content: "Use 'find_by(id:)' instead of 'where(id:).first' to avoid loading entire relation",
  memory_type: "instruction",
  ticket_id: 100
)

fact - Learned Information

# Technical fact
store_memory(
  content: "Tailwind CSS v4 uses @import instead of @tailwind directives. CSS config moved to @plugin block in CSS file itself",
  memory_type: "fact",
  ticket_id: 89
)

# Project-specific fact
store_memory(
  content: "This project uses Tinkered workflow: Workers implement → Reviewers audit → Orchestrators coordinate",
  memory_type: "fact",
  ticket_id: 98
)

summary - Completed Work

# After completing a ticket
store_memory(
  content: "Ticket #89: Implemented daisyUI theme polish. 3 themes configured (winter, dracula, corporate). All AVO cards converted to daisyUI components. Created theme switcher controller and partial.",
  memory_type: "summary",
  ticket_id: 89
)

# After completing a phase
store_memory(
  content: "Epic #79 Phase 1 Complete: daisyUI installed, theme configured, base components converted. Ready for Phase 2: Advanced features.",
  memory_type: "summary",
  ticket_id: 79
)

Search Patterns

Search by Ticket

search_memory(query: "ticket #89", limit: 10)
search_memory(query: "#114", limit: 5)

Search by Type

search_memory(memory_type: "decision", limit: 20)
search_memory(memory_type: "error", query: "daisyUI", limit: 10)

Search by Keyword

search_memory(query: "authentication", limit: 10)
search_memory(query: "Stacked PRs", limit: 5)
search_memory(query: "AVO card", limit: 10)

Search by Ticket Context

# Get all memories for current ticket
search_memory(ticket_id: 114, limit: 20)

Workflow Integration

Standard Work Pattern

# 1. Start: Search for context
search_memory(query: "ticket #114", limit: 5)
search_memory(query: "memory skill pattern", limit: 5)

# 2. Work: Implement solution
# ... code changes ...

# 3. Learn: Store discoveries
store_memory(content: "Memory skills use same directory structure as git-workflow", memory_type: "fact", ticket_id: 114)

# 4. Complete: Store summary
store_memory(content: "Created memory skill with search/store guidance, memory types, examples", memory_type: "summary", ticket_id: 114)

Memory Quality Checklist

When storing memory, ensure:

• ✅ Specific: Contains actionable details
• ✅ Typed: Uses correct memory_type
• ✅ Context: Includes ticket_id or metadata
• ✅ Searchable: Uses keywords others might search for
• ✅ Unique: Doesn't duplicate existing memories

When searching memory:

• ✅ Before starting: Check for prior work
• ✅ When stuck: Look for similar solutions
• ✅ Before deciding: Check prior decisions
• ✅ Use filters: Apply memory_type, ticket_id when relevant