TracineHQ / open source for Claude Code
Open-source tools that make AI coding agents safer, more transparent, and easier to trust in production workflows.
Stdlib-only safety hooks for Claude Code. Validates bash commands, protects credentials, enforces git discipline, scopes subagents, and guards protected files.
/plugin marketplace add TracineHQ/plugins/plugin install guard@tracineSQLite-backed analytics CLI for Claude Code sessions. Indexes session logs, full-text search across transcripts, tool-call analytics, and atomic snapshot/restore.
/plugin marketplace add TracineHQ/plugins/plugin install convo@tracineDid the eval get better, or is that noise? A Claude Code skill and CLI that sizes the runs before you test, gates CI on the verdict, and says "can't tell yet" when the data can't support a call. Zero-dependency JS and scipy-backed Python, cross-validated in CI.
/plugin marketplace add TracineHQ/plugins/plugin install eval-kit@tracinePlanned: LLM-assisted triage for static analysis findings. Not built yet.
Guard's hooks fire before the tool call. Block dangerous shell commands, credential exfiltration, and edits to protected files before they execute, not after.
Every Guard decision writes to JSONL. Every session writes to SQLite. Full-text search across what your agent actually did, not what it claimed it did.
A 2.7-point bump on four runs is not a result. eval-kit sizes the runs before you test, gates CI on the verdict, and says "can't tell yet" when the data can't support a call. The math is checked against scipy in CI.