card

/session-memory:card — Create a Knowledge Card

You are helping the user create a knowledge card for the team's collective memory.

Steps

1. Parse arguments: Extract title from the first argument. Optional flags:

   • --type TYPE: decision | constraint | gotcha | runbook | onboarding (default: gotcha)
   • --importance N: 0-3 (default: 2)
   • --visibility V: team | private (default: team)

2. Gather context: Look at the current session's recent work:

   • What files were recently edited (check git diff or recent tool usage)
   • What errors were encountered
   • What decisions were made

3. Draft the card: Create a markdown file with YAML frontmatter:

---
id: mem_YYYYMMDD_XXXXXX
title: <user's title>
type: <type>
files: [<relevant files>]
tags: [<inferred tags>]
importance: <importance>
status: active
visibility: <visibility>
confidence: med
created: <today>
author: <user@host>
refs: []
---

## TL;DR
<1-2 sentence summary of the knowledge>

## Details
<Detailed explanation with context>

## Evidence
- <today> session-<id>: Created manually

4. Secret scanning: Before writing the file, check the content for secrets:

   • AWS keys (AKIA...), API keys (sk-..., ghp_...), Bearer tokens
   • Database URLs (postgres://, mysql://, mongodb://)
   • If found: mask with REDACTED and warn the user

5. Save: Write to .claude/memory/cards/mem_YYYYMMDD_XXXXXX.md

   • Generate unique ID using date + random hex
   • Confirm creation to the user
   • If --visibility private, mention it won't be shared via git

Card Types Guide

Type	When to use	Example
decision	Recording why a technical choice was made	"Chose PostgreSQL over MongoDB for ACID compliance"
constraint	Documenting immutable limitations	"API rate limit is 100 requests/min"
gotcha	Warning about non-obvious pitfalls	"req.user can be null before auth middleware"
runbook	Step-by-step operational procedures	"Database migration: run X, then Y, then Z"
onboarding	New team member guides	"Dev environment setup: install A, configure B"

Important Rules

• Card files go in .claude/memory/cards/
• Always run secret scanning on the content before writing
• Use today's date for the created field
• Generate a unique ID: mem_YYYYMMDD_ + 6 random hex chars
• Tags should be auto-inferred from the content + file paths
• If the user doesn't provide a title, ask for one