opencode-plugin-reliability-loop

name	opencode-plugin-reliability-loop
description	Use when iterating OpenCode plugin behavior and you need scored artifacts, baseline comparison, and regression-aware next-step guidance.

OpenCode Plugin Reliability Loop

Overview

Use this workflow to tune plugin behavior with fast, repeatable scoring.

Loop

1. Generate a fresh session export JSON.
2. Run npm run reliability:loop -- --export <session-export.json>.
3. Review latest report in .opencode/reliability/runs/<run-id>/report.md.
4. Check trend and baseline in .opencode/reliability/scoreboard.json.
5. Fix highest-impact reason code and rerun.

Strict Labels

• post_nudge_user_correction (strong false-positive evidence)
• post_no_nudge_user_prompt_to_continue (strong false-negative evidence)

Both labels use a strict 1-turn window for user feedback.

Soft Gate

• Fail on ACP smoke failure.
• Fail on strong score regression.
• Warn on mild score regression.
• Pass when no strong regression exists.

Artifacts

• .opencode/reliability/runs/<run-id>/run.json
• .opencode/reliability/runs/<run-id>/events.jsonl
• .opencode/reliability/runs/<run-id>/labels.json
• .opencode/reliability/runs/<run-id>/scores.json
• .opencode/reliability/runs/<run-id>/diff-vs-baseline.json
• .opencode/reliability/runs/<run-id>/report.md
• .opencode/reliability/scoreboard.json