memory-record

Memory Record Skill

When the user shares a fact, decision, preference, pattern, or learned lesson, save it to the structured memory system immediately.

Activation Triggers

Trigger when:

• User says "remember to...", "I prefer...", "learned that...", "note that..."
• User mentions a decision or approach that worked well
• User discovers a pattern, workaround, or gotcha
• Context suggests something non-obvious or hard-won knowledge
• User explicitly asks you to remember something

File Selection

Route to correct memory file:

1. memory.md (index only)
   • Do NOT add content here—this is auto-managed
2. general.md — Cross-project facts, preferences, environment
   • User workflow preferences ("I prefer terse responses")
   • Global tool setup ("RTK installed at X")
   • Cross-project patterns ("always verify before destructive ops")
3. domain/{topic}.md — Domain-specific knowledge
   • React patterns, database design, testing strategies
   • File path: domain/{topic}.md where topic is 1-2 words (kebab-case)
   • Examples: domain/authentication.md, domain/api-design.md
4. tools/{tool}.md — Tool configs, CLI patterns, workarounds
   • Git workflows, CLI tricks, editor config
   • Examples: tools/git.md, tools/bash.md, tools/postgres.md

Entry Format

Each entry has exactly 3 lines:

**[YYYY-MM-DD]** what you learned — why it matters

Example:

**[2026-05-01]** Always run tests before pushing to main — saved us from catching broken merges in CI
**[2026-05-01]** RTK proxy rewrites git commands transparently — saves 60-90% tokens on dev operations

Rules:

• Date in [YYYY-MM-DD] format
• "What" is 1 sentence (what you learned/decided/prefer)
• "Why" is optional but encouraged (helps future-you decide if still relevant)
• No markdown formatting except the date brackets
• One entry per line

Steps

1. Evaluate: Is this worth remembering? (It should be non-obvious or hard-won)
2. Categorize: Pick the right file (general/domain/tools)
3. Create if needed: If file doesn't exist, create it with empty frontmatter and entries
4. Append entry: Add new entry to the file with date, fact, and why
5. Update memory.md: Add/update one-line index entry for the file
6. Confirm to user: Show what was saved and where

memory.md Index Format

Maintain one-line entries per file:

- [general.md](general.md) — cross-project preferences and global setup
- [domain/authentication.md](domain/authentication.md) — user auth patterns, JWT vs session tradeoffs
- [tools/git.md](tools/git.md) — git workflows, cherry-pick patterns, merge strategies

Update whenever you modify a memory file. Keep sorted: general first, then domains (alphabetical), then tools (alphabetical).

Example Workflow

User says: "I prefer terse responses with no trailing summaries"

Skill executes:

1. Categorize → general.md (cross-project preference)
2. Create/open general.md
3. Append: **[2026-05-01]** Prefer terse responses, no trailing summaries — saves tokens and respects user flow
4. Update memory.md index
5. Confirm: "Saved to general.md: terse response preference"

User says: "React components re-render when object refs change inline"

Skill executes:

1. Categorize → domain/react.md (domain-specific)
2. Create/open domain/react.md if not exist
3. Append: **[2026-05-01]** Inline object props create new ref each render = re-render — wrap in useMemo or move outside component
4. Update memory.md index
5. Confirm: "Saved to domain/react.md: inline object re-render gotcha"