Memory Management Skill

Purpose

Give Claude persistent memory across sessions using native file-based storage — no external CLI or dependencies required. Memory is stored as structured markdown files and searched semantically by reading and reasoning over them.

Memory Structure

Memory lives in a memory/ folder at the root of your current workspace:

memory/
  clients.md
  skill-patterns.md
  signals.md

Three namespaces:

File	What Goes In It
`memory/clients.md`	Client names, ICPs, preferences, past work, tone
`memory/skill-patterns.md`	What worked when building or running skills, reusable patterns
`memory/signals.md`	High-scoring topics from monitoring, trending skill gaps

When to Use

Before starting any task:

Read the relevant memory file(s)
Check if there's existing context that applies

After completing any task:

Store what worked (pattern, decision, insight)
Update existing entries rather than duplicating

How to Store a Memory

First time using a namespace? Use the Write tool to create the file. After that, use Edit to append new entries.

If a file doesn't exist yet, create it with a header:

# [Namespace Name]

Then add entries below.

Use the Edit tool to append to an existing file. Format:

## [Topic or Client Name]
- [Key fact or pattern]
- [Date added: YYYY-MM-DD]

Example — client context:

## Acme Corp
- ICP: B2B SaaS CTOs at Series A companies replacing legacy tools
- Preferred tone: direct, no buzzwords
- Past work: lead research for their DevOps product (March 2026)
- Date added: 2026-03-09

Example — skill pattern:

## Lead Research — What Works
- Always use WebSearch, never generate companies from training data
- CSV export at the end increases session value significantly
- Mode selector (own business vs client) reduces back-and-forth
- Date added: 2026-03-09

Example — signal pattern:

## High-Signal Topics (March 2026)
- "Code review automation" trending on Twitter, 3 weeks consistent
- "Proposal writing assistant" gap confirmed — no skillsmp match
- Date added: 2026-03-09

How to Search Memory

Read the relevant file and reason over it:

Read memory/clients.md → find entries matching the current client or project
Read memory/skill-patterns.md → find patterns relevant to the current task type
Read memory/signals.md → find recent signal trends relevant to the current topic

If the file doesn't exist or is empty: proceed without memory — nothing to apply yet. This is expected on first use.

Decision Logic

START TASK
  → Read relevant memory file
  → Apply any matching context or patterns
  → Proceed with task

END TASK
  → Did you learn something reusable?
    YES → Append to the relevant memory file
    NO  → Skip

Best Practices

Check memory before starting — avoid re-learning what's already known
Store after completing — one line is enough if that's all there is
Update existing entries rather than creating duplicates
Keep entries concise — memory files should be scannable, not essays
Use the signals.md file to feed insights back into the skill pipeline

memory-management