mega-code-skill-enhance

Run on-demand A/B evaluation of a mega-code skill, review results in an HTML viewer, collect user feedback, and produce an enhanced version of the skill. The host agent (you) handles test generation, grading, and enhancement; isolated A/B completions run via subprocess.

Setup

MEGA_DIR="$(cd "${CLAUDE_SKILL_DIR}/../.." && pwd)"
set -a && . ~/.local/share/mega-code/.env 2>/dev/null && set +a
uv run --directory "$MEGA_DIR" python -m mega_code.client.check_auth

If the auth check fails (non-zero exit), show the output to the user and stop.

Detect which agent you are — set EVAL_AGENT so the A/B runner uses the same agent:

• If you are Claude Code, set EVAL_AGENT=claude
• If you are Codex, set EVAL_AGENT=codex
• If unsure, omit it (auto-detection will be used)