jarvis-reload

JaRVIS Session Reload

Reload your identity and memories mid-session. This is useful after reflections update your memories, or when you need a fresh read of your context.

Note: On platforms with session-start hooks (Claude Code, Cursor), identity loads automatically. Copilot and OpenCode have hooks for marker tracking but cannot inject context — run this skill at the start of each session. On other platforms (Antigravity, etc.), run this skill manually at the start of each session.

Step 1: Resolve JaRVIS data directory

Run JARVIS_DIR=$(bash <skill-path>/scripts/resolve-dir.sh) to set JARVIS_DIR.

If the resolved directory doesn't exist, inform the user that JaRVIS hasn't been set up yet and offer to run /jarvis-init to scaffold it.

If it exists, proceed.

Step 2: Run pending data-dir migrations

Bring the data dir up to date with the current plugin version:

bash <skill-path>/../jarvis-migrate/scripts/migrate.sh "$JARVIS_DIR"

(Replace <skill-path>/.. with the path to the jarvis-migrate skill — it's a sibling of jarvis-reload.)

• Silent exit 0 = data dir already current; proceed silently.
• Output beginning with JaRVIS data dir migrated v<old> → v<new> followed by bullets = migrations ran. Surface the changelog to the user before continuing.
• Non-zero exit = stop. Report the error to the user; do not load identity/memories until the migration issue is resolved.

Step 3: Load your identity

Read $JARVIS_DIR/IDENTITY.md. This is who you are. Internalize it:

• This is who you are, developed over real experiences over time.

Step 4: Load your memories

Use /jarvis-search to retrieve your consolidated memories:

• Search type: memory
• Section: Consolidated

This extracts only the curated, deduplicated knowledge from all memory files. Internalize these memories — they inform how you work with this user and project.

If the consolidated results are empty or contain only placeholder text (e.g., "No consolidated … yet"), search the Recent section instead — early memories live there before consolidation kicks in.

If you need more detail on a specific topic, read individual memory files in $JARVIS_DIR/memories/ directly.

Step 5: Report

Give a brief, natural summary — not a data dump. Something like (but doesn't have to be!):

"I'm [name], v[version]. Last session we [brief summary of most recent journal entry]. I'm an agent that has done [key decision] and you prefer [key preference]. Ready to work."

Keep it concise. The point is to show you have context, not to recite everything you know.

Note on platform memory

Some platforms have their own auto-memory systems (e.g., Claude Code's ~/.claude/projects/ MEMORY.md) that run separately and handle incidental observations. Don't duplicate those into JaRVIS memories. JaRVIS memories are for deliberate, reflected-on knowledge — things that came out of the reflection process with context and rationale. If your platform's memory already captured something small like a build command or file path, there's no need to also store it in JaRVIS memories.