Vercel Plugin Eval

Plugin Eval

Launch real Claude Code sessions with the plugin installed, monitor debug logs in real-time, and verify every hook fires correctly with proper dedup.

DO NOT (Hard Rules)

• DO NOT use claude --print or -p — hooks don't fire, no files created
• DO NOT use --dangerously-skip-permissions
• DO NOT create projects in /tmp/ — always use ~/dev/vercel-plugin-testing/
• DO NOT manually wire hooks or create settings.local.json — use npx add-plugin
• DO NOT set CLAUDE_PLUGIN_ROOT manually
• DO NOT use bash -c in WezTerm — use /bin/zsh -ic
• DO NOT use full path to claude — use the x alias
• DO NOT write eval scripts — do everything as Bash tool calls in the conversation

Copy the exact commands below. Do not improvise.

Quick Start

Always append a timestamp to directory names so reruns don't overwrite old projects:

# 1. Create test dir & install plugin (with timestamp)
TS=$(date +%Y%m%d-%H%M)
SLUG="my-eval-$TS"
mkdir -p ~/dev/vercel-plugin-testing/$SLUG
cd ~/dev/vercel-plugin-testing/$SLUG
npx add-plugin https://github.com/vercel/vercel-plugin -s project -y

# 2. Launch session via WezTerm
wezterm cli spawn --cwd /Users/<user>/dev/vercel-plugin-testing/$SLUG -- /bin/zsh -ic \
  "unset CLAUDECODE; VERCEL_PLUGIN_LOG_LEVEL=debug x '<PROMPT>' --settings .claude/settings.json; exec zsh"

# 3. Find debug log (wait ~25s for session start)
find ~/.claude/debug -name "*.txt" -mmin -2 -exec grep -l "$SLUG" {} +

What to Monitor

Hook firing (all 8 registered hooks)

LOG=~/.claude/debug/<session-id>.txt

# SessionStart (3 hooks)
grep "SessionStart.*success" "$LOG"

# PreToolUse skill injection
grep -c "executePreToolHooks" "$LOG"        # total calls
grep -c "provided additionalContext" "$LOG"  # injections

# UserPromptSubmit
grep "UserPromptSubmit.*success" "$LOG"

# PostToolUse validate + shadcn font-fix
grep "posttooluse-validate.*provided" "$LOG"
grep "PostToolUse:Bash.*success" "$LOG"

# SessionEnd cleanup
grep "SessionEnd" "$LOG"

Dedup correctness (the key metric)

TMPDIR=$(node -e "import {tmpdir} from 'os'; console.log(tmpdir())" --input-type=module)
CLAIMDIR="$TMPDIR/vercel-plugin-<session-id>-seen-skills.d"

# Claim files = one per skill, atomic O_EXCL
ls "$CLAIMDIR"

# Compare: injections should equal claims
inject_meta=$(grep -c "skillInjection:" "$LOG")
claims=$(ls "$CLAIMDIR" 2>/dev/null | wc -l | tr -d ' ')
echo "Injections: $((inject_meta / 3)) | Claims: $claims"

skillInjection: appears 3x per actual injection in the debug log (initial check, parsed, success). Divide by 3.

PostToolUse validate quality

Look for real catches — API key bypass, outdated models, wrong patterns:

grep "VALIDATION" "$LOG" | head -10

Scenario Design

Describe products and features, never name specific technologies. Let the plugin infer which skills to inject. Always end prompts with: "Link the project to my vercel-labs team so we can deploy it later. Skip any planning and just build it. Get the dev server running."

Coverage targets by scenario type

Hard-to-trigger skills (8 of 44)

These need explicit technology references in the prompt because agents don't naturally reach for them:

• ai-elements — say "use the AI Elements component registry"
• v0-dev — say "generate components with v0"
• vercel-firewall — say "use Vercel Firewall for rate limiting"
• marketplace — say "publish to the Vercel Marketplace"
• geist — say "install the geist font package"
• json-render — name files components/chat-*.tsx

Coverage Report

Write results to .notes/COVERAGE.md with:

1. Session index — slug, session ID, unique skills, dedup status
2. Hook coverage matrix — which hooks fired in which sessions
3. Skill injection table — which of the 44 skills triggered
4. Dedup stats — injections vs claims per session
5. Issues found — bugs, pattern gaps, validation findings

Cleanup

rm -rf ~/dev/vercel-plugin-testing