Menu
Manage your long-term memory via MEMORY.md
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Manages cron jobs for this Right Agent agent via MCP tools. Creates, updates, and deletes cron specs stored in the agent database. The Rust runtime handles scheduling and execution automatically. Use when the user mentions cron jobs, scheduled tasks, reminders, one-shot tasks, or recurring tasks.
Use when you verified a reusable procedure this turn (non-trivial multi-step work, concrete gotchas, or you just corrected a wrong approach) and want to save or fix a rightx-* skill, or when the user explicitly asks to save, remember, or fix one. Not for one-off tasks, unverified guesses, or trivial single-step work.
Use BEFORE calling any mcp__right__composio__* tool — especially COMPOSIO_SEARCH_TOOLS, COMPOSIO_MULTI_EXECUTE_TOOL, or COMPOSIO_REMOTE_WORKBENCH — and whenever a request involves Notion, Gmail, Google Calendar, Google Drive, Slack, GitHub, Linear, or any other service routed through Composio. Sessions that skip this skill burn 300-500K context tokens on repeated tool searches and inline payloads ($200+ per session observed in cache-write cost). Read once per session, on resume, and any turn that mentions a Composio-fronted service.
Finds and adds MCP servers for this Right Agent agent. Searches for OAuth-capable endpoints first (Claude Code / Codex integration docs), falls back to API-key endpoints. All management goes through the user's Telegram dashboard MCP view — the agent never handles credentials directly. Use when the user asks to add, connect, or set up an MCP server or integration.
Manage your long-term memory powered by Hindsight
Inspects this agent's own past reasoning by reading the conversation-history JSONL files Claude Code writes inside the sandbox. Use when the user asks "why did you...", "what were you thinking when...", or to debug a wrong decision the agent made in an earlier turn or session. Reads thinking blocks, tool calls, and tool results to reconstruct prior reasoning. Filesystem-only — no MCP calls, no DB.
| name | right-memory |
| description | Manage your long-term memory via MEMORY.md |
Your long-term memory is stored in MEMORY.md in your home directory.
This file is automatically injected into your system prompt at the start
of each interaction (truncated to 200 lines).
When the user says "remember", "save this", or "don't forget", first classify what kind of persistent fact it is. The word "remember" means "persist this", not "write to MEMORY.md".
TOOLS.mdUSER.mdSOUL.mdIDENTITY.mdAfter this routing, memory is the fallback for facts with no better home.
Use Claude Code's built-in Edit and Write tools to manage MEMORY.md:
Things that don't have a home in your other files (TOOLS.md,
IDENTITY.md, SOUL.md, USER.md):
USER.mdRoute these to the correct file instead of writing to MEMORY.md:
TOOLS.mdUSER.mdIDENTITY.md / SOUL.mdMEMORY.md is re-read as context on every turn. Imperative phrasing ("Always do X") gets interpreted as a directive and can override the user's current request.
"User prefers pytest-xdist for parallel tests" ✓"Always run tests with pytest -n 4" ✗"API foo returns 422 when inputis used instead ofarguments" ✓"Use arguments for API foo" ✗ (this is a rule — goes in TOOLS.md)MEMORY.md is truncated to 200 lines in your prompt. Periodically: