Hardstop is a local-only Claude Code plugin. This document describes what data it collects and how it's handled.
Hardstop only processes and stores data locally on your machine:
| Data | Purpose | Storage |
|---|---|---|
| Command text | Safety analysis | Logged to ~/.hardstop/audit.log |
| Working directory | Context for analysis | Logged to audit.log |
| Verdict (ALLOW/BLOCK) | Decision record | Logged to audit.log |
| Timestamp | Audit trail | Logged to audit.log |
- No personal information
- No conversation history
- No API keys or credentials
- No telemetry or analytics
- No data sent to external servers
All data is stored locally in ~/.hardstop/:
~/.hardstop/
├── state.json # Plugin enabled/disabled state
├── skip_next # Temporary one-time bypass flag
└── audit.log # Command decision log (JSON-lines)
Hardstop sends no data to external servers.
The only network activity is the optional LLM analysis layer, which uses the local Claude CLI (claude --print). This runs within your existing Claude subscription and is subject to Anthropic's privacy policy.
audit.loggrows over time as commands are analyzed- You can delete
~/.hardstop/at any time to remove all stored data - No automatic data expiration or cleanup
Hardstop uses:
- Claude CLI (optional) — For LLM-based semantic analysis of edge cases
- No other third-party services
You can:
- View all stored data:
cat ~/.hardstop/audit.log - Delete all data:
rm -rf ~/.hardstop/ - Disable logging: Not currently supported (audit log is always written)
Updates will be documented in the repository changelog.
For privacy questions: contact@clarity-gate.org
Last updated: 2025-01-17