claude-code-internals

<constraints>
- **Output limit**: Always use `| head -N` (SIGTRAP risk from session bloat)
- **Relative paths**: Use `scripts/`, `references/` (not absolute paths)
</constraints>

# Claude Code Internals Explorer

Procedures for analyzing Claude Code's standalone binary.

## Quick Start

> **Note**: All script paths are relative to this skill's directory (where SKILL.md resides).

**Delegate all binary exploration to subagent:**

```
Task tool (subagent_type: Explore)
Prompt: "Run scripts/find_installation.sh to get binary path.
        Then search for [keyword] using: strings $BINARY | grep -E '[pattern]' | head -50
        Return: version, matching lines with context (-B2 -A2)."
```

The subagent handles token-heavy strings output; main context receives only summarized findings.

## Zero-Context Mode (Recommended)

**Problem**: Subagent output is recorded in session logs. Binary analysis can cause session file bloat → SIGTRAP crashes.

**Solution**: Write prompt to temp file, then execute via Headless CLI wrapper. Script internals are not logged; only final stdout is recorded.

### Workflow

1. **Write prompt** to temp file (supports multiline)
2. **Execute** `scripts/analyze-binary.sh <prompt_file> [allowed_tools]`

### Script Interface

| Argument | Description | Default |
|----------|-------------|---------|
| `$1` | Prompt file path (required) | - |
| `$2` | Allowed tools (comma-separated) | `Bash,Read,Glob,Grep` |

The script injects `BINARY_PATH` automatically.

### Example

```bash
# Use unique ID in filename for parallel independence
PROMPT_FILE="/tmp/internals_prompt_$$.txt"

# 1. Write prompt to temp file
cat > "$PROMPT_FILE" << 'EOF'
Search the binary for beta headers.

Instructions:
1. Use: strings "$BINARY_PATH" | grep -E "anthropic-beta|20[0-9]{2}-[0-9]{2}" | head -30
2. List all matching headers with their apparent status
3. Note any patterns suggesting feature flags
EOF

# 2. Execute with tool restriction
scripts/analyze-binary.sh "$PROMPT_FILE" "B