Hindsight Memory Workflow

Guide for effectively using Hindsight memory bank during development sessions.

CRITICAL: Automatic Behavior

As an AI assistant, you MUST use Hindsight memory proactively:

Always Check Memory BEFORE:

• •Making architectural recommendations
• •Suggesting technology choices
• •Proposing design patterns
• •Answering "how should we..." questions

Use the recall skill to search memory.

Always SAVE to Memory AFTER:

• •User makes a decision with reasoning ("Let's use X because Y")
• •Bug is solved ("Found it! The problem was...")
• •Tradeoff is discussed ("We're sacrificing X for Y")
• •Convention is established ("From now on...")
• •User says "remember", "important"

Use the retain skill to store to memory.

Trigger Phrases (SAVE immediately when you hear):

• •"Let's use...", "We decided...", "The reason is..."
• •"Remember that...", "From now on...", "Important:..."
• •"The solution was...", "We chose X over Y..."

When to Access Memory

Recall Memory (Search)

Access memory bank when:

1. •

   Before making architectural decisions

   • •"Let me check if we've discussed this approach before..."
   • •Query: architectural patterns, design decisions for [component]
2. •

   When choosing technologies or libraries

   • •"I should verify our previous technology choices..."
   • •Query: tech stack, library choices, [technology name]
3. •

   When encountering similar problems

   • •"This looks familiar, checking memory..."
   • •Query: bugs, issues, problems with [component/feature]
4. •

   When the user asks about past decisions

   • •Explicit: "What did we decide about X?"
   • •Implicit: "Why is it done this way?"
5. •

   When context seems missing

   • •"There might be context I'm not aware of..."
   • •Query: requirements, constraints, [feature area]

Retain Memory (Store)

Store information when:

1. •

   Architectural decisions are made

   code

   Example: "We decided to use event-driven architecture for the notification system because it allows better scalability and decoupling"
   Context: architecture
   

2. •

   Technology choices are finalized

   code

   Example: "Chose Redis for caching over Memcached due to better data structure support and persistence options"
   Context: tech-stack
   

3. •

   Complex bugs are solved

   code

   Example: "Memory leak in WebSocket handler was caused by event listeners not being cleaned up on disconnect"
   Context: bugs
   

4. •

   Tradeoffs are explicitly discussed

   code

   Example: "Sacrificed some type safety by using 'any' in the legacy adapter to maintain backward compatibility"
   Context: tradeoffs
   

5. •

   Important lessons are learned

   code

   Example: "Integration tests caught the race condition that unit tests missed - always test concurrent scenarios"
   Context: lessons
   

6. •

   Requirements or constraints are established

   code

   Example: "API response time must stay under 200ms per product requirements"
   Context: requirements
   

Reflect (Analysis)

Use reflection when:

1. •

   Evaluating new approaches against existing patterns

   • •"Given our past decisions, should we adopt microservices?"
2. •

   Looking for patterns in past decisions

   • •"What architectural patterns have worked well for us?"
3. •

   Making recommendations based on history

   • •"What approach would align with our established conventions?"

Memory Categories

Best Practices

For Storing Memories

1. •

   Be specific and include reasoning

   • •Bad: "Use PostgreSQL"
   • •Good: "Use PostgreSQL for the users service because we need ACID transactions for financial data"
2. •

   Include context about alternatives considered

   • •"Chose X over Y because of Z"
3. •

   Store at the moment of decision, not later

   • •Context is freshest immediately after discussion
4. •

   Use appropriate categories

   • •Helps with future recall accuracy

For Retrieving Memories

1. •

   Be specific in queries

   • •Bad: "database"
   • •Good: "database choice for user authentication"
2. •

   Use appropriate budget

   • •low for quick lookups of recent/obvious info
   • •mid for general queries
   • •high for comprehensive analysis
3. •

   Cross-reference with current context

   • •Memory might be outdated; verify relevance

Action Skills

For execution details, the following action skills handle all CLI interactions:

• •retain skill — saves content to memory bank (handles bank_id, CLI flags, heredoc)
• •recall skill — searches memory bank (handles bank_id, budget, max-tokens)
• •reflect skill — gets AI analysis from memory bank (handles bank_id, budget, max-tokens)