supacortex

Original：🇺🇸 English
Translated

Personal memory layer — save bookmarks and conversation summaries using the Supacortex CLI. Use when the user says "save to cortex", "save to supacortex", "save this session", or asks to recall past conversations.

3installs
Sourcemonorepo-labs/skills
Added on

NPX Install

npx skill4agent add monorepo-labs/skills supacortex

Tags

Translated version includes tags in frontmatter
cli-toolpersonal-memory-layerbookmark-managementconversation-managementai-agent-tool

SKILL.md Content

View Translation Comparison →

Supacortex — CLI Skill

Supacortex is a personal memory layer. Two CLI commands: bookmarks and conversations.

Setup

bash
npm i -g @supacortex/cli
scx login

1. Bookmarks (
scx bookmarks
)

Save and search bookmarks (links, YouTube videos).

Commands

List bookmarks

bash
scx bookmarks list [--search "<query>"] [--type <tweet|link|youtube>] [--limit <n>] [--offset <n>] [--pretty]

Add a bookmark

bash
scx bookmarks add <url> [--pretty]

Get a bookmark by ID

bash
scx bookmarks get <id> [--pretty]

Delete a bookmark

bash
scx bookmarks delete <id> [--pretty]

When to use bookmarks

  • User asks to save a link or YouTube video
  • User wants to search their saved content
  • When you need to reference previously saved URLs

2. Conversations (
scx conversation
)

Save summaries of AI chat sessions. Every conversation has a tier that determines its depth.

Tiers

TierWhen to useContent format
brief
Throwaway queries, quick lookupsSingle sentence: "Asked about JSON parsing in Bun"
summary
Most working sessionsMarkdown — 3-8 bullet points covering what was discussed, decided, and found
detailed
Deep sessions with architectural decisions, research findingsMarkdown — full structured document with headings, reasoning, code snippets, follow-ups

Commands

Save a conversation

bash
scx conversation add "<content>" --tier <brief|summary|detailed> [--title "<title>"] [--metadata '<json>'] [--pretty]
The 
--tier
flag is required. It maps to memory types: 
conversation_brief
, 
conversation_summary
, 
conversation_detailed
.
Examples:
bash
# Brief — one sentence
scx conversation add "Helped debug CORS issue in Hono API" --tier brief

# Summary — bullet points
scx conversation add "- Set up memory table with tsvector search
- Added triggers for auto search vector generation
- Created API routes for CRUD
- Decided on hybrid schema approach" \
  --tier summary \
  --title "Memory system setup" \
  --metadata '{"source": "claude-code"}'

# Detailed — full document
scx conversation add "## Memory Architecture Decision..." --tier detailed --title "Memory layer brainstorm"

List conversations

bash
scx conversation list [--search "<query>"] [--tier <brief|summary|detailed>] [--limit <n>] [--offset <n>] [--pretty]

Get a conversation by ID

bash
scx conversation get <id> [--pretty]

Update a conversation

bash
scx conversation update <id> [--title "<title>"] [--content "<content>"] [--tier <tier>] [--metadata '<json>'] [--pretty]

Delete a conversation

bash
scx conversation delete <id> [--pretty]

When to save conversations

Save when the user says:
  • "save to cortex" / "save to supacortex"
  • "save this session" / "remember this"
  • "log this conversation"

When to recall conversations

Pull past conversations when the user says:
  • "check my conversation about X"
  • "pull the X conversation"
  • "get the summary for X"
  • "what did we work on last time?"

Metadata

Metadata is freeform JSON passed via 
--metadata
. The AI decides what to store. Common fields:
  • source
    — where this was captured ("claude-code", "chatgpt", "opencode")
  • tags
    — array of topic tags
  • project
    — which project the conversation was about (e.g. "supacortex", "supalytics")
  • category
    — topic area: "project", "life", "general", "learning", "work"

Consistent verbs across all commands

Both commands (
bookmarks
, 
conversation
) support: 
list
, 
add
, 
get
, 
delete
.
conversation
also supports: 
update
.
All commands output JSON by default (optimized for AI agents). Use 
--pretty
for human-readable output.