Remember Skill

A meta-skill that helps you capture and store reminders, preferences, and workflows as persistent guidance. When you say "remember to [do something]", this skill helps determine the best way to encode that reminder so it applies to future work.

When to Use This Skill

Manual Invocation

Use /remember when you want to create a new reminder about:

•A recurring workflow or process
•A preference or guideline you want applied to future work
•An automation or check you should run before starting a task
•A pattern you've noticed that should be codified

Automatic Recognition

Claude should proactively suggest this skill when you say things like:

•"Remember to [action] next time [trigger]"
•"Make sure we [action] before [situation]"
•"I should [action] whenever [condition]"

Workflow

Phase 1: Clarify the Reminder

First, understand what you want to remember by asking:

•
What should happen? (the action/behavior to remember)
- •What is the specific action or check you want to happen?
- •Is it something automated, manual, or a mix?
•
When should it happen? (the trigger/context)
- •What triggers this? (before task start, during certain workflows, on PR reviews, etc.)
- •Is it a one-time thing or recurring?
•
How important/frequent is this?
- •How often will this come up?
- •What's the impact if it's forgotten?
•
Who does it apply to?
- •Is this just for you or for anyone working on this project?
- •Should it apply to all projects or just this one?

Phase 2: Decide Storage Method

Use this decision tree to pick the best approach:

code

Is this a workflow/tool/command that needs to be runnable on demand?
├─ YES → Create a NEW SKILL
│  └─ Examples:
│     - "remember to check Jira before starting" → skill that lists Jira issues
│     - "remember to run my validation checks" → skill that bundles multiple commands
│     - "remember to ask for approval" → skill that guides a workflow
│
├─ NO → Does this enhance or extend an existing skill?
│  ├─ YES → UPDATE THE EXISTING SKILL
│  │  └─ Examples:
│  │     - Add a new command to gh-pr-comments skill
│  │     - Extend jira skill with new functionality
│  │
│  └─ NO → Does this apply to all projects or this specific one?
│     ├─ ALL PROJECTS → Update AGENTS.md (global instructions)
│     │  └─ Examples:
│     │     - "remember to always use the Explore agent for broad searches"
│     │     - "remember to check task list before starting work"
│     │     - "remember to document all architectural decisions"
│     │
│     └─ THIS PROJECT → Update CLAUDE.md (project instructions)
│        └─ Examples:
│           - "remember to run Gazelle before mocks in Go"
│           - "remember to use testify suite for Go tests"

Phase 3: Execute with Approval

•
Show the user what you'll create
- •Display the new skill/updates you'll make
- •Explain why this approach was chosen
- •Ask for approval before writing anything
•
Get explicit approval
- •User must say yes (or equivalent) before you write files
- •This prevents accidentally overwriting instructions
•
Write and commit
- •Create the skill/update the file
- •Create a git commit with clear message
- •Confirm completion

Examples

Example 1: Create a New Skill

User says: "Remember to check our Jira backlog before starting any feature work"

Decision: This is a workflow that needs to run on demand → Create new skill

Action: Create ~/development/toolbox/ai/skills/jira-backlog-check/SKILL.md with:

•Description of the skill
•How to invoke it
•What commands it runs (likely jira skill calls + filtering)
•Clear output format

Example 2: Update Existing Skill

User says: "Remember to also check for Jira subtasks, not just parent tickets"

Decision: This enhances the existing jira skill → Update existing skill

Action: Add guidance to ~/development/toolbox/ai/skills/jira/SKILL.md about subtask handling

Example 3: Update Project Instructions

User says: "Remember to always run compass workspace clean --expunge before building after long breaks"

Decision: This is a build system best practice for this project → Update CLAUDE.md

Action: Add to /Users/nathaniel.morihara/development/urbancompass/CLAUDE.md under Build/Test Commands section

Example 4: Update Global Instructions

User says: "Remember to always ask users to clarify ambiguous requirements before planning implementation"

Decision: This is a general guideline for all projects → Update AGENTS.md

Action: Add to ~/development/toolbox/AGENTS.md under guidelines section

Implementation Tips

When creating new skills:

•Make the skill name descriptive and lowercase with hyphens
•Include clear "when to use" section explaining the trigger
•Provide example commands and outputs
•Reference related skills if applicable

When updating CLAUDE.md or AGENTS.md:

•Preserve existing content structure
•Add your reminder in the most appropriate section
•Keep explanations concise but clear
•Link to related sections if needed

When updating existing skills:

•Don't remove or break existing content
•Add new capabilities at the end of sections
•Update examples if they become incomplete
•Maintain the existing tone and style

Approval Workflow

code

User triggers /remember or mentions "remember to X"
         ↓
Phase 1: Ask clarifying questions
         ↓
Phase 2: Determine which file/skill to create/update
         ↓
Phase 3: Show user the proposed changes (BEFORE writing)
         ↓
User approves or provides feedback
         ↓
Write changes and create commit
         ↓
Confirm completion

Critical: Always show the user what you plan to write before writing it.