| name | load-brain |
| description | Load the essential Second Brain (large) into context. Your core original thinking: journals, inventions, predictions, writing, health, lists. Light trim removes redundant metadata; keeps Summary, Keywords, Connections, Original Text. Excludes external sources, bulk fiction, transcripts, code/builds, and docs already loaded via rules. Use when user says "load brain", "load everything", "full context", "deep context". |
| allowed-tools | ["Bash","Read"] |
Load Brain
Reads your core intellectual output into context via a pre-built snapshot file.
Stub notice: scripts/build_brain_snapshot.py is not yet implemented. Fill it in
(see SETUP.md) before /load-brain will work end-to-end. Once implemented, this skill
runs exactly as described below.
Context Budget (1M model)
| Component | Approximate size |
|---|
| Snapshot content | large |
| Read tool overhead | ~55K tokens |
| System context | ~75K tokens |
| Conversation | remaining headroom |
| Total | ~1,000K |
The exact snapshot size depends on your content. Run python3 scripts/build_brain_snapshot.py
to see a breakdown after implementing the script.
What Gets Loaded
Light trim: ## Metadata, ## Classification, ## Source, ## Status removed during build.
Kept: Title, Summary, Project Connections, Connections, Keywords, Original Text.
| Category | Description |
|---|
| Inventions | Original invention and idea notes |
| Journal Intellectual | Intellectual / philosophical entries |
| Journal Personal | Personal reflective entries |
| Writing SciFi | Original science fiction writing |
| Predictions | Forward-looking notes and forecasts |
| Writing AllElse | Essays, posts, miscellaneous writing |
| Health | Health observations and tracking |
Configure ESSENTIAL_DIRS in the script to match your own content folders (see SETUP.md).
What's Excluded (and why)
| Category | Reason |
|---|
| Orientation Docs (key files) | Already loaded via .claude/rules/ symlinks |
| Projects/ docs | STATE_OF_SECOND_BRAIN (via rules) covers project status |
| External sources / third-party | Not your original thinking |
| AI-generated drafts | Unreviewed; noise in a synthesis context |
| Bulk fiction (large sub-folder) | Token-heavy; load separately when needed |
| Raw transcripts / vault | Massive token footprint; rarely needed whole |
| Reference / To_Study | External references, learning queue |
| Code, builds, media, binary | Not prose content |
Process
Step 1: Regenerate Snapshot
python3 scripts/build_brain_snapshot.py
This builds two files (paths relative to repo root):
.claude/brain_snapshot.md — the concatenated content (gitignored).claude/brain_load_instructions.md — exact Read parameters per batch
Step 2: Read the Instruction File
Read(file_path="<repo-root>/.claude/brain_load_instructions.md")
The instruction file is short (~75 lines). It contains the exact Read calls grouped into batches,
generated from the actual line count of the snapshot.
Step 3: Execute Batch 1
Fire ALL Read calls listed under ## Batch 1 as parallel Read calls.
Use the EXACT file_path, offset, and limit values from the instruction file.
Step 4: Execute Batch 2
Fire ALL Read calls listed under ## Batch 2 as parallel Read calls.
Step 5: Execute Batch 3
Fire ALL Read calls listed under ## Batch 3 as parallel Read calls.
Step 6: Execute Additional Batches (if any)
If the instruction file has more than 3 batches (the brain grew), execute each additional
batch the same way.
Step 7: Confirm
Report to user with the snapshot stats printed by the build script:
- Total files and tokens loaded
- Remaining conversation headroom
- Any orientation docs already in context via rules (skip re-reading those)
CRITICAL RULES
These exist because the Read tool has a hard 10,000 token per call limit.
650 lines = ~8,125 tokens. This is safe. 1,000 lines = ~12,500 tokens. This FAILS.
- NEVER use limit > 650. Every Read call MUST use
limit=650 or less.
- NEVER combine batches. Execute Batch 1, wait for ALL reads to complete, then Batch 2, etc.
- NEVER modify offsets or limits. Use the EXACT values from
brain_load_instructions.md.
- NEVER improvise a different chunking strategy. The instruction file is generated from
the actual file size and is always correct.
- NEVER stop between batches to ask the user. Load everything, then report.
- NEVER recalculate chunk sizes. The instruction file already did the math.
- If a single Read call fails, retry that specific call once. If it fails again, skip it and continue.
- The instruction file is the source of truth. If this SKILL.md text conflicts with the
instruction file, the instruction file wins.
Freshness
- Snapshot + instructions are regenerated each time
/load-brain runs (Step 1).
- Also regenerate after significant content additions (e.g., during maintenance runs).
FILE: markers inside the snapshot show the exact source path for traceability.
When to Use
- Starting a deep analysis session ("tell me about my patterns")
- Cross-referencing ideas across the entire brain
- Finding connections between distant topics
- Preparing for a writing or synthesis session that needs full context
