~ / rigs / naman10parikh / agentbench

naman10parikh/agentbench

Lighthouse for coding-agent harnesses: score your setup out of 100.

↗ GitHub ★ 0 MIT updated 28d ago project Claude CodeCodexGemini CLI Claude plugin
share on X
ARCHETYPE
Orchestrator
A bench of specialised subagents. The main agent mostly delegates.
CONTEXT TAX · EVERY TURN
up to ~7.2k tokens
Claude Code loads MCP tools on demand, so the real cost is usually lower.
Heavy · median rig: 2.3k · breakdown
GUARDRAILS
3/5
Blocks destructive commands · Pre-tool screening hook · No YOLO mode · details
Quality 84/100 Guardrails v2

Copy this rig

# review before running: this installs third-party code
$ claude mcp add agentbench -- npx -y -p github:naman10parikh/agentbench agentbench-mcp
$ npx degit naman10parikh/agentbench/.claude ./rig-agentbench  # inspect, then merge into .claude/

MCP servers are added to Claude Code at local scope; env vars are shown as YOUR_… placeholders — we never store values. Files are fetched with degit into a separate folder so you can review before merging.

$ claude mcp add agentbench -- npx -y -p github:naman10parikh/agentbench agentbench-mcp
$ npx degit naman10parikh/agentbench/.claude/skills/agent-runtime .claude/skills/agent-runtime
$ npx degit naman10parikh/agentbench/.claude/skills/anthropic-frontend-design .claude/skills/anthropic-frontend-design
$ npx degit naman10parikh/agentbench/.claude/skills/architect .claude/skills/architect
$ npx degit naman10parikh/agentbench/.claude/skills/deep-think .claude/skills/deep-think
$ npx degit naman10parikh/agentbench/.claude/skills/design-systems .claude/skills/design-systems
$ npx degit naman10parikh/agentbench/.claude/skills/ecosystem-reference .claude/skills/ecosystem-reference
$ npx degit naman10parikh/agentbench/.claude/skills/fractal-delegation .claude/skills/fractal-delegation
$ npx degit naman10parikh/agentbench/.claude/skills/harness-review .claude/skills/harness-review
$ npx degit naman10parikh/agentbench/.claude/skills/heartbeat .claude/skills/heartbeat
$ npx degit naman10parikh/agentbench/.claude/skills/integrate-resources .claude/skills/integrate-resources
$ npx degit naman10parikh/agentbench/.claude/skills/llm-council-critique .claude/skills/llm-council-critique
$ npx degit naman10parikh/agentbench/.claude/skills/loop-create .claude/skills/loop-create
$ npx degit naman10parikh/agentbench/.claude/skills/mcp-setup .claude/skills/mcp-setup
$ npx degit naman10parikh/agentbench/.claude/skills/memory-compression .claude/skills/memory-compression
$ npx degit naman10parikh/agentbench/.claude/skills/memory .claude/skills/memory
$ npx degit naman10parikh/agentbench/.claude/skills/model-routing .claude/skills/model-routing
$ npx degit naman10parikh/agentbench/.claude/skills/oss-project-scaffold .claude/skills/oss-project-scaffold
$ npx degit naman10parikh/agentbench/.claude/skills/playwright-cli .claude/skills/playwright-cli
$ npx degit naman10parikh/agentbench/.claude/skills/repo-scout .claude/skills/repo-scout
$ npx degit naman10parikh/agentbench/.claude/skills/self-improve .claude/skills/self-improve
$ npx degit naman10parikh/agentbench/.claude/skills/skill-creator .claude/skills/skill-creator
$ npx degit naman10parikh/agentbench/.claude/skills/skill-routing .claude/skills/skill-routing
$ npx degit naman10parikh/agentbench/.claude/skills/sprint .claude/skills/sprint
$ npx degit naman10parikh/agentbench/.claude/skills/test-visual .claude/skills/test-visual
$ npx degit naman10parikh/agentbench/.claude/skills/troubleshoot .claude/skills/troubleshoot
$ npx degit naman10parikh/agentbench/.claude/skills/ux-audit .claude/skills/ux-audit
$ npx degit naman10parikh/agentbench/skills/add-benchmark-task .claude/skills/add-benchmark-task
$ npx degit naman10parikh/agentbench/skills/benchmark-harness .claude/skills/benchmark-harness
$ npx degit naman10parikh/agentbench/skills/memory-search .claude/skills/memory-search
$ npx degit naman10parikh/agentbench/skills/sandbox-run .claude/skills/sandbox-run
$ curl -fsSL --create-dirs -o .claude/agents/architect.md https://raw.githubusercontent.com/naman10parikh/agentbench/main/.claude/agents/architect.md
$ curl -fsSL --create-dirs -o .claude/agents/bench-runner.md https://raw.githubusercontent.com/naman10parikh/agentbench/main/.claude/agents/bench-runner.md
$ curl -fsSL --create-dirs -o .claude/agents/code-reviewer.md https://raw.githubusercontent.com/naman10parikh/agentbench/main/.claude/agents/code-reviewer.md
$ curl -fsSL --create-dirs -o .claude/agents/loop-auditor.md https://raw.githubusercontent.com/naman10parikh/agentbench/main/.claude/agents/loop-auditor.md
$ curl -fsSL --create-dirs -o .claude/agents/performance-analyzer.md https://raw.githubusercontent.com/naman10parikh/agentbench/main/.claude/agents/performance-analyzer.md
$ curl -fsSL --create-dirs -o .claude/agents/research-agent.md https://raw.githubusercontent.com/naman10parikh/agentbench/main/.claude/agents/research-agent.md
$ curl -fsSL --create-dirs -o .claude/agents/security-reviewer.md https://raw.githubusercontent.com/naman10parikh/agentbench/main/.claude/agents/security-reviewer.md
$ curl -fsSL --create-dirs -o .claude/agents/task-author.md https://raw.githubusercontent.com/naman10parikh/agentbench/main/.claude/agents/task-author.md
$ curl -fsSL --create-dirs -o .claude/agents/test-writer.md https://raw.githubusercontent.com/naman10parikh/agentbench/main/.claude/agents/test-writer.md

Merge into .claude/settings.json (project) or ~/.claude/settings.json (user). Hook commands reference scripts in the source repo — copy those too.

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [
          {
            "type": "command",
            "command": "node \"$CLAUDE_PROJECT_DIR/.claude/hooks/block-dangerous-commands.cjs\""
          }
        ]
      },
      {
        "matcher": "Read|Edit|Write|Bash",
        "hooks": [
          {
            "type": "command",
            "command": "node \"$CLAUDE_PROJECT_DIR/.claude/hooks/protect-secrets.cjs\""
          }
        ]
      }
    ]
  }
}

MCP servers (1)

serversourceest. tokens
github:naman10parikh/agentbench · "agentbench" npm 2.5k

Skills (30)

Subagents (9)

architect
model: sonnet
System architect agent for evaluating trade-offs, designing APIs, reviewing schemas, and planning multi-service integration.
bench-runnerScores a Claude Code harness with agentbench and explains the result. Use when someone wants a number for their setup or wants to compare two setups. Runs the free static score first and the paid benc
code-reviewer
model: sonnet
Reviews code changes for quality, security, and alignment with Energy platform architecture.
agent-auditor
model: sonnet
Audits agent definitions (currently .energy format) for completeness, token efficiency, and harness quality.
performance-analyzer
model: sonnet
Analyzes code for performance bottlenecks, memory leaks, bundle size issues, and optimization opportunities.
research-agent
model: haiku
Deep research agent for investigating topics, reading docs, searching repos, and summarizing findings. Optimized for breadth.
security-reviewer
model: sonnet
Reviews code changes for security vulnerabilities (OWASP Top 10, secrets exposure, injection attacks, auth issues).
task-authorAdds or fixes a graded coding task in the agentbench benchmark, end to end. Use when the benchmark needs a new kind of task or a task's prompt, files or rubric is wrong. Works in tasks/index.ts and sr
test-writer
model: sonnet
Generates comprehensive test suites for Energy platform modules.

Hooks (3)

eventmatcherruns
SessionStart*bash "$CLAUDE_PROJECT_DIR/.claude/hooks/session-start-context.sh"
PreToolUseBashnode "$CLAUDE_PROJECT_DIR/.claude/hooks/block-dangerous-commands.cjs"
PreToolUseRead|Edit|Write|Bashnode "$CLAUDE_PROJECT_DIR/.claude/hooks/protect-secrets.cjs"

Slash commands (16)

/compress/deps-audit/dx-gha/feature-dev/handoff/integrate-resources/perf-check/preserve/recover/research/resume/security-scan/start/status/validate/wrap-up

Similar rigs

copied ✓