RegtraceAgent Skill
Agent skill for teaching AI agents about the regtrace CLI
skills/regtrace/ is an agent skill that teaches AI coding agents how to
use the regtrace CLI. Agents load it automatically when a user asks about
LLM output quality, golden sets, or regtrace commands.
Installation
npx skills add decimozs/regtrace/skills/regtraceThis registers the skill with your agent harness (OpenCode, Claude Code, Cursor, Zed, etc.). The agent then loads it automatically when you ask about LLM quality, golden sets, or regtrace commands.
Contents
| File | Purpose |
|---|---|
SKILL.md | Trigger phrases, quick-start, golden set format, score interpretation, common tasks |
cli.md | Full CLI reference: commands, flags, config YAML, golden set format, env vars |
Coverage
The skill covers:
- Project scaffolding and configuration
- Golden set YAML format (single-turn and RAG)
- Evaluation metrics — factuality, format, tone, regression
- Quality gates and pass/fail criteria
- All CLI commands and flags
- CI integration patterns
- Debugging failing test cases
- Judge provider configuration
- Run comparison via
history --diff
Agent harness integration
Type /regtrace init inside an agent harness (OpenCode, Claude Code, Cursor,
Zed, Aider, etc.) to bootstrap regtrace without writing config by hand. The
agent scans your project for LLM signals — prompts, client calls, RAG chains,
tools, and API routes — then generates golden sets and config wired to your
actual code.
See skills/regtrace/SKILL.md for the full workflow.