docs(showcase): add bug-fix-benchmark example for SWE-bench style evaluation

[christso]

Summary

Add examples/showcase/bug-fix-benchmark/ — a SWE-bench style showcase that compares coding agent performance with and without engineering plugins on identical bug fix tasks.

Uses target-level hooks (from #1095) to configure each plugin variant via execution.targets in the eval file.

Closes #919

What It Does

Evaluates the same bug fix task across four configurations:

Target	Plugin	Workspace
claude-baseline	None	baseline settings
claude-superpowers	obra/superpowers	superpowers plugin
claude-compound	EveryInc/compound-engineering-plugin	compound plugin
claude-agent-skills	addyosmani/agent-skills	agent-skills plugin

Metrics compared: tokens consumed, time to complete, fix correctness.

How It Works

A single claude base target is defined in targets.yaml. The eval file uses target-level hooks (execution.targets) so each variant runs a before_each hook that copies the appropriate .claude/settings.json into the workspace:

execution:
  targets:
    - name: claude-baseline
      use_target: claude
      hooks:
        before_each:
          command: ["bash", "../scripts/setup-variant.sh", "baseline"]
    - name: claude-superpowers
      use_target: claude
      hooks:
        before_each:
          command: ["bash", "../scripts/setup-variant.sh", "superpowers"]
    # ...

Files Added

examples/showcase/bug-fix-benchmark/
├── .agentv/targets.yaml           # Base claude target + grader targets
├── evals/bug-fixes.eval.yaml      # Test cases + target hooks per variant
├── workspaces/                    # Plugin config templates (copied by hooks)
│   ├── baseline/                  # bypassPermissions, no plugins
│   ├── superpowers/               # + superpowers plugin
│   ├── compound/                  # + compound-engineering plugin
│   └── agent-skills/              # + agent-skills plugin
├── scripts/
│   ├── setup-variant.sh           # Target hook: copy variant config into workspace
│   ├── mock-agent.sh              # Test harness without API keys
│   ├── setup-plugins.sh           # Install plugins into workspaces
│   └── import-swebench.sh         # Import SWE-bench dataset instances
└── README.md

How to Run

# All variants (runs all 4 from execution.targets)
agentv eval evals/bug-fixes.eval.yaml --workers 3

# Specific variants only
agentv eval evals/bug-fixes.eval.yaml \
  --target claude-baseline,claude-superpowers --workers 2

# Compare results
agentv compare <baseline-results> <plugin-results>

Test Plan

• bun run build — passes
• bun run typecheck — passes
• bun run lint — passes
• bun run test — 2157 tests pass (1642 core + 67 eval + 448 cli)
• bun run validate:examples — 56/56 valid
• Dry-run with --dry-run --workers 1 — all 4 targets resolve hooks, clone repo, copy correct settings.json
• Verified workspace slots contain correct plugin configs per variant

🤖 Generated with Claude Code

[cloudflare-workers-and-pages]

Deploying agentv with    Cloudflare Pages

Latest commit:	a85f0b4
Status:	⚡️  Build in progress...

View logs