@claudemini/shit-cli 1.4.0 → 1.6.0

package/DESIGN_PHILOSOPHY.md

@@ -1,138 +0,0 @@
-# shit-cli Design Philosophy
-
-## Core Vision
-
-### 1. Human-AI Interaction Memory System
-Long-term memory storage for human-AI interactions, not just temporary logs. Each session is analyzed for semantic meaning — what was the intent, what changed, what's the risk.
-
-### 2. Code Review Bot Data Support
-Provide reliable, structured data to support code review automation. The bot doesn't need to parse raw events — it gets pre-classified changes, risk assessments, and review hints.
-
-## Architecture
-
-```
-Human ↔ AI (Claude Code)
-         ↓ (hooks)
-    Event Ingestion (log.js)
-         ↓
-    Semantic Extraction (extract.js)
-         ↓
-    Session State (session.js) + Reports (report.js)
-         ↓
-    Memory System (.shit-logs + index.json)
-         ↓
-    Code Review Bot / Human Queries
-```
-
-### Three-Layer Architecture
-
-1. **Ingestion Layer** — `log.js` reads stdin, parses events, dispatches
-2. **Intelligence Layer** — `extract.js` classifies intent, changes, risk
-3. **Storage Layer** — `session.js` manages state, `report.js` generates outputs
-
-## Semantic Model
-
-### Intent Extraction
-From user prompts, extract:
-- **Goal**: What the user is trying to accomplish
-- **Type**: bugfix, feature, refactor, debug, test, docs, etc.
-- **Scope**: Which domains are involved (auth, api, database, etc.)
-
-### Change Extraction
-From tool events, extract:
-- **Files**: Categorized (source, test, config, doc, script, infra, deps, migration)
-- **Operations**: What was done to each file (read, write, edit)
-- **Commands**: Categorized (test, build, git, deploy, install, lint, database)
-
-### Session Classification
-Combining intent + changes:
-- **Type**: Dominant activity category
-- **Risk**: Assessment based on file count, config changes, test coverage
-- **Review Hints**: Actionable flags for code review bots
-
-## File Structure
-
-```
-.shit-logs/
-├── index.json               # Cross-session index (file history, session types)
-└── <session-id>/
-    ├── events.jsonl          # Raw hook events (complete history)
-    ├── state.json            # Incremental processing state
-    ├── summary.json          # Bot data interface (v2 schema)
-    ├── summary.txt           # Human-readable semantic report
-    ├── prompts.txt           # User prompts with timestamps
-    └── metadata.json         # Lightweight session metadata
-```
-
-## Key Design Decisions
-
-### 1. Rule-Based Classification (No AI Dependency)
-Intent and change classification use simple pattern matching rules. This ensures:
-- Zero latency — runs during hook processing
-- Zero cost — no API calls
-- Predictable — deterministic output
-- Offline — works without network
-
-### 2. Cross-Session Index
-The `index.json` file enables:
-- File history queries ("how often was this file changed?")
-- Session type filtering ("show all bugfix sessions")
-- Risk tracking over time
-- Bot can query without scanning all sessions
-
-### 3. summary.json v2 Schema
-Upgraded from v1 (statistics-only) to v2 (semantic):
-- **v1**: event counts, file lists, tool usage
-- **v2**: intent, type, risk, review_hints, categorized changes/commands
-
-### 4. Incremental Processing
-State is maintained incrementally in `state.json`. Each event updates the state, and reports are regenerated from the latest state. This means:
-- Fast processing per event (~5ms)
-- Reports always reflect current session state
-- No need to re-read events.jsonl
-
-### 5. Best-Effort Shadow Branches
-Git shadow branches and index updates are best-effort — failures don't affect core logging. This ensures hooks never block Claude Code.
-
-## Bot Integration Patterns
-
-### Direct File Read
-```javascript
-const summary = JSON.parse(fs.readFileSync('.shit-logs/<id>/summary.json'));
-// Use summary.review_hints for automated review decisions
-```
-
-### Cross-Session Query
-```javascript
-const index = JSON.parse(fs.readFileSync('.shit-logs/index.json'));
-// Find all sessions that touched a specific file
-const sessions = index.file_history['src/auth/service.ts'];
-// Filter by type or risk
-const risky = index.sessions.filter(s => s.risk === 'high');
-```
-
-### CLI Query (for humans)
-```bash
-shit query --file=src/auth/service.ts  # File history
-shit query --type=bugfix --recent=5    # Recent bugfixes
-shit query --risk=high --json          # High-risk as JSON
-```
-
-## Comparison with `.entire`
-
-| Aspect | `.entire` | `.shit-logs` |
-|--------|-----------|--------------|
-| **Data Source** | Full transcript | Hook events |
-| **Intelligence** | Raw messages | Semantic extraction |
-| **Bot Interface** | Parse messages | Structured JSON (v2) |
-| **Cross-Session** | No | Index with file history |
-| **Review Hints** | No | Tests, risk, coverage |
-| **Purpose** | Conversation memory | Operation intelligence |
-
-## Future Enhancements
-
-- AI-powered intent extraction (optional, when available)
-- Diff-level change tracking (what exactly changed in each edit)
-- Session linking (detect related sessions across time)
-- Anomaly detection (unusual patterns in session activity)
-- Bot API endpoint (serve session data over HTTP)

package/QUICKSTART.md

@@ -1,109 +0,0 @@
-# Quick Start Guide
-
-## Installation
-
-```bash
-# Clone or download shit-cli
-cd ~/Desktop/shit-cli
-
-# Install globally
-npm link
-
-# Or add to PATH
-echo 'export PATH="$HOME/Desktop/shit-cli/bin:$PATH"' >> ~/.zshrc
-source ~/.zshrc
-```
-
-## Setup (One-time)
-
-```bash
-# Navigate to your project
-cd /path/to/your/project
-
-# Initialize hooks
-shit init
-```
-
-This creates `.claude/settings.json` with all necessary hooks.
-
-## Verify Installation
-
-```bash
-# Check if shit is available
-shit help
-
-# Should output:
-# shit-cli - Session-based Hook Intelligence Tracker
-# Usage: shit <command> [options]
-# ...
-```
-
-## Usage
-
-Once initialized, hooks will automatically log events to `~/.shit-logs/`.
-
-### View your sessions
-
-```bash
-# List all sessions
-shit list
-
-# View specific session
-shit view <session-id>
-```
-
-### Clean up old logs
-
-```bash
-# Preview what will be deleted
-shit clean --days=7 --dry-run
-
-# Actually delete
-shit clean --days=7
-```
-
-## Troubleshooting
-
-### Command not found: shit
-
-If `npm link` doesn't work, use the full path:
-
-```bash
-node ~/Desktop/shit-cli/bin/shit.js init
-```
-
-Or create an alias:
-
-```bash
-echo 'alias shit="node ~/Desktop/shit-cli/bin/shit.js"' >> ~/.zshrc
-source ~/.zshrc
-```
-
-### Hooks not working
-
-1. Check if `.claude/settings.json` exists in your project
-2. Verify hooks are registered: `cat .claude/settings.json`
-3. Restart Claude Code
-
-### Logs not appearing
-
-1. Check log directory: `ls -la ~/.shit-logs/`
-2. Verify hooks are being triggered (check Claude Code output)
-3. Test manually:
-   ```bash
-   echo '{"session_id":"test-123","hook_event_name":"Test"}' | shit log test
-   shit list
-   ```
-
-## Uninstall
-
-```bash
-# Remove global link
-npm unlink -g shit-cli
-
-# Remove logs
-rm -rf ~/.shit-logs
-
-# Remove hooks from project
-# Edit .claude/settings.json and remove shit-related hooks
-```