Unsevered Memory - Workflow Skill

You are operating with persistent memory. This skill defines how to maintain context across sessions.

Memory Architecture

Two Sources of Truth

Location	Type	Purpose	Update Frequency
.claude/memory/	Dynamic	Session state, decisions	Every session
.ai/	Static	Architecture, patterns	When patterns emerge

File Purposes

File	Content	When to Update
context.md	Current state, next steps	End of session
scratchpad.md	Live session operations	During session
decisions.md	Architectural choices	When decisions made
.ai/core/	Tech stack, architecture	When they change
.ai/patterns/	Reusable solutions	After 3+ uses

---

Session Workflow

On Session Start

1. Memory hook automatically loads context.md
2. Review the current state and task
3. Check scratchpad.md for unfinished work
4. Reference .ai/ for patterns and architecture

During Session

Write to scratchpad.md as you work:

## Session: 2024-01-15 14:30

### Operations
- [14:31] Read context.md - last session worked on auth
- [14:32] User asked to fix login bug
- [14:35] Found issue in validateToken() at src/auth.ts:142
- [14:40] Fixed: was comparing wrong field

### Findings
- Token validation was checking user.id instead of user.uid
- Only affects users created after migration

### Decisions
- Keep backward compatibility by checking both fields

### Blockers
(none)

### Next Steps
- Add regression test for token validation

Update .ai/ when patterns emerge:

After implementing the same solution 3+ times, add it to .ai/patterns/.

On Session End

1. Hook archives scratchpad to sessions/YYYY-MM-DD.md
2. Update context.md with:
   • What was accomplished
   • Current state
   • Next steps
3. Append architectural decisions to decisions.md

---

Live Documentation Updates

When to Update .ai/

During your session, proactively update .ai/ when you encounter:

Trigger	Location	Example
Pattern used 3+ times	.ai/patterns/	Error handling, API structure
Architecture change	.ai/core/architecture.md	New service layer
Stack change	.ai/core/technology-stack.md	New dependency
Workflow improvement	.ai/workflows/	Better test process

How to Update

Do not wait for session end. Update as you go:

1. Notice a pattern emerging
2. Add/update relevant .ai/ file immediately
3. Log the update in scratchpad
4. Continue working

Example scratchpad entry:

### Operations
- [14:35] Implemented error handling in 3 endpoints
- [14:40] Pattern detected - created .ai/patterns/error-handling.md
- [14:45] Applied pattern to remaining endpoints

---

Decision Tree: Where Does This Go?

Is this information:

[Specific to today's work?]
    YES --> scratchpad.md

[An architectural decision?]
    YES --> decisions.md (also update .ai/ if changes architecture)

[A reusable pattern?]
    YES --> .ai/patterns/

[About the tech stack?]
    YES --> .ai/core/technology-stack.md

[Current project state?]
    YES --> context.md

---

File Templates

context.md

# Project Context

## Current State
[What is the current state of the project/feature?]

## Current Task
[What are we working on right now?]

## Last Session
- Date: YYYY-MM-DD
- Accomplished: [summary]
- Stopped at: [where we left off]

## Next Steps
1. [First thing to do]
2. [Second thing to do]

## Active Branches
- feature/xxx - [purpose]

## Notes
[Any context the next session needs]

scratchpad.md

# Scratchpad

Session: YYYY-MM-DD HH:MM

## Operations
- [HH:MM] Action taken

## Findings
- What was discovered

## Decisions
- Choices made and rationale

## Blockers
- Issues encountered

## Next Steps
- What to do if session ends

decisions.md

# Decision Log

## YYYY-MM-DD: [Decision Title]

**Context**: [Why was this decision needed?]

**Options Considered**:
1. Option A - pros/cons
2. Option B - pros/cons

**Decision**: [What was chosen]

**Rationale**: [Why this option]

**Consequences**: [What this means going forward]

---

Enforcement

The UserPromptSubmit hook displays memory state on every prompt:

[Memory] Task: Fix auth bug | Scratchpad: 24 lines | .ai/ updated: 2024-01-15

This reminder survives context compaction. Use it to stay oriented.

---

Commands

/orchestrate

Activates orchestrator mode for complex tasks. The orchestrator:

1. Reads all memory files
2. Breaks task into subtasks
3. Delegates to specialized agents
4. Updates memory after each step
5. Never loses context

Use for multi-step tasks requiring persistent state management.