[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-94782":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":5,"homepage":8,"htmlUrl":8,"language":9,"languages":8,"totalLinesOfCode":8,"stars":10,"forks":11,"watchers":11,"openIssues":12,"contributorsCount":13,"subscribersCount":13,"size":13,"stars1d":13,"stars7d":13,"stars30d":12,"stars90d":13,"forks30d":13,"starsTrendScore":13,"compositeScore":14,"rankGlobal":8,"rankLanguage":8,"license":15,"archived":16,"fork":16,"defaultBranch":17,"hasWiki":18,"hasPages":16,"topics":19,"createdAt":8,"pushedAt":8,"updatedAt":20,"readmeContent":21,"aiSummary":22,"trendingCount":13,"starSnapshotCount":13,"syncStatus":23,"lastSyncTime":24,"discoverSource":25},94782,"commerce-agent-bench","harukiseller-droid\u002Fcommerce-agent-bench","harukiseller-droid",null,"Python",105,101,4,0,46.43,"MIT License",false,"main",true,[],"2026-08-24 04:01:22","# Commerce Agent Bench\n\nReproducible skills, regression fixtures and evaluation workflows for coding agents maintaining ecommerce codebases.\n\n`commerce-agent-bench` gives coding agents a shared protocol for reviewing WooCommerce, Shopify, and static ecommerce projects without inventing product facts or silently turning assumptions into claims. It combines portable agent instructions with deterministic regression fixtures that can run locally and in CI.\n\n## Primary jobs\n\n- Ecommerce pull-request review.\n- Commerce fact safety for buyer-facing and structured-data claims.\n- WooCommerce\u002FShopify regression detection.\n\n## Why this exists\n\nAI agents are useful at code review, migrations, SEO fixes, schema work, and ecommerce maintenance, but commerce code has failure modes that generic coding benchmarks rarely cover:\n\n- hard-coded prices, reviews, or inventory claims;\n- duplicate WooCommerce hooks and theme overrides;\n- broken product-page accessibility;\n- missing SEO metadata and structured-data regressions;\n- fabricated product facts in generated copy;\n- platform-specific fixes that ignore runtime product data.\n\nThis repository turns those problems into reusable skills, review recipes, and small eval fixtures.\n\n## What is included\n\n- **Portable agent skills** in `.agents\u002Fskills\u002F` for ecommerce PR review, commerce fact safety, product-page review, technical SEO, schema, WooCommerce, and accessibility. Pinterest generation is an experimental extra under `extras\u002F`.\n- **Deterministic scanner** in `src\u002Fcommerce_agent_bench\u002F` for fast regression checks.\n- **Reproducible evals** in `evals\u002F` with intentionally broken fixtures and expected rule IDs.\n- **Safe examples** for WooCommerce, Shopify Liquid, and static storefronts.\n- **GitHub Actions CI** to run tests and evals on every pull request.\n- **Agent protocol** in `AGENTS.md` so Codex, Claude Code, Cursor, Gemini CLI, and similar tools can follow the same evidence-first workflow.\n\n## Quick start\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fharukiseller-droid\u002Fcommerce-agent-bench.git\ncd commerce-agent-bench\npython -m venv .venv\nsource .venv\u002Fbin\u002Factivate\npython -m pip install -e . pytest\ncommerce-agent-bench evals\u002Ffixtures\u002Fbroken-product-page --format text\npytest -q\npython scripts\u002Frun_evals.py\npython scripts\u002Fvalidate_benchmarks.py\n```\n\nExpected eval summary:\n\n```text\nPASS product-page-basics\nPASS schema-integrity\nPASS woocommerce-hook-regression\nPASS accessibility-regression\nPASS seo-regression\nPASS shopify-product-card-accessibility\nPASS woocommerce-product-summary-hook\nPASS hardcoded-price\nPASS fake-review-count\nPASS fake-stock\nPASS unsafe-template-escaping\nPASS missing-price-currency\nPASS canonical-conflict\nPASS duplicate-product-schema\nPASS fabricated-shipping\nPASS fabricated-dimensions\nPASS invalid-product-jsonld\nPASS shopify-static-price\nPASS empty-cta\n\n19\u002F19 eval cases passed\n```\n\nThe current deterministic benchmark summary, including case-level results, is in [`docs\u002Fbenchmarks\u002F`](docs\u002Fbenchmarks\u002F). Regenerate it with:\n\n```bash\npython scripts\u002Frun_benchmarks.py\n```\n\n## Codex benchmark status\n\nThese scenarios are packaged for real Codex execution. No Codex runtime has been run from this repository yet, so every scenario is explicitly `NOT RUN`.\n\n| Benchmark | Status | Evidence |\n|---|---|---|\n| Hardcoded price | NOT RUN | `benchmarks\u002Fcodex\u002F001-hardcoded-price\u002F` |\n| Unsupported rating | NOT RUN | `benchmarks\u002Fcodex\u002F002-fake-rating\u002F` |\n| Duplicate WooCommerce hook | NOT RUN | `benchmarks\u002Fcodex\u002F003-duplicate-wc-hook\u002F` |\n| Shopify runtime data | NOT RUN | `benchmarks\u002Fcodex\u002F004-shopify-runtime-data\u002F` |\n| Missing product facts | NOT RUN | `benchmarks\u002Fcodex\u002F005-fabricated-product-fact\u002F` |\n\nCodex benchmark execution: `NOT RUN`. The validator rejects incomplete or unsupported result artifacts.\n\nThe optional `.github\u002Fworkflows\u002Fcodex-review.yml` workflow is manual, opt-in, and currently `NOT VERIFIED`. It does not run in normal CI, call an API, or claim Codex execution. Enablement requires an explicit repository variable and a configured secret, but an actual adapter must still be implemented and tested before it can review a pull request.\n\n## Use with a coding agent\n\nPoint the agent at `AGENTS.md`, then ask it to run one of the workflows:\n\n```text\nAudit this WooCommerce product template using the `.agents\u002Fskills\u002Fproduct-page-audit`\nand `.agents\u002Fskills\u002Fwoocommerce-code-review` skills. Separate FACT, INFERENCE, and UNKNOWN.\nDo not invent product facts. Include file paths and line evidence.\n```\n\nThe core contract is tool-agnostic. Agent-specific files should stay thin and refer back to the shared protocol instead of duplicating it.\n\n## Output contract\n\nAgent findings should use this shape:\n\n```text\nstatus: FACT | INFERENCE | UNKNOWN\nseverity: low | medium | high | critical\ncategory: seo | schema | accessibility | commerce-data | platform | content\nlocation: path:line\nfinding: concise description\nevidence: exact code or observed behavior\nrisk: why it matters\nrecommended_patch: smallest safe fix\nverification: how to prove the fix worked\n```\n\n## Current deterministic rules\n\n| Rule | Severity | Purpose |\n|---|---:|---|\n| `HTML_IMG_ALT_MISSING` | medium | Detect image tags without `alt` |\n| `HTML_BUTTON_NAME_MISSING` | medium | Detect empty unnamed buttons |\n| `SEO_TITLE_MISSING` | high | Detect HTML documents without `\u003Ctitle>` |\n| `SEO_META_DESCRIPTION_MISSING` | medium | Detect missing meta descriptions |\n| `SCHEMA_FAKE_RATING` | high | Flag suspicious hard-coded rating values |\n| `SCHEMA_FAKE_REVIEW_COUNT` | high | Flag suspicious hard-coded review counts |\n| `WOOCOMMERCE_DUPLICATE_HOOK` | high | Flag duplicate product-summary hook registration |\n| `UNSAFE_HARDCODED_PRICE` | medium | Flag likely hard-coded commerce prices |\n| `UNSAFE_HARDCODED_STOCK` | medium | Flag likely hard-coded stock or availability |\n| `HTML_UNESCAPED_TEMPLATE_OUTPUT` | high | Flag unescaped Liquid product output |\n| `SCHEMA_PRICE_CURRENCY_MISSING` | medium | Flag schema prices without currency evidence |\n| `SEO_CANONICAL_CONFLICT` | high | Flag multiple canonical links |\n| `SCHEMA_DUPLICATE_PRODUCT` | high | Flag duplicate Product entities |\n| `UNSAFE_HARDCODED_SHIPPING` | medium | Flag hard-coded shipping or delivery times |\n| `UNSAFE_HARDCODED_DIMENSIONS` | medium | Flag hard-coded product dimensions |\n| `SCHEMA_INVALID_JSONLD` | high | Flag invalid JSON-LD fixtures |\n| `SHOPIFY_STATIC_PRODUCT_PRICE` | medium | Flag static Shopify prices |\n\nThese checks are intentionally small and explainable. They are not a replacement for platform linters, browser tests, or human review.\n\n## Repository layout\n\n```text\ncommerce-agent-bench\u002F\n├── AGENTS.md\n├── PROTOCOL.md\n├── README.md\n├── APPLICATION.md\n├── .agents\u002F\n│   └── skills\u002F\n├── extras\u002F\n│   └── pinterest-content-generator\u002F\n├── recipes\u002F\n├── src\u002Fcommerce_agent_bench\u002F\n├── tests\u002F\n├── evals\u002F\n│   ├── fixtures\u002F\n│   ├── expected\u002F\n│   └── manifest.json\n├── examples\u002F\n├── benchmarks\u002F\n│   └── codex\u002F\n├── scripts\u002F\n├── docs\u002F\n└── .github\u002F\n```\n\n## Design principles\n\n1. **Evidence before conclusions.** Agents must cite code, rendered behavior, or supplied product data.\n2. **No fabricated commerce facts.** Unknown shipping, price, material, inventory, dimensions, review counts, or guarantees remain `UNKNOWN`.\n3. **Runtime data over hard-coded copy.** Product-specific values should come from the platform or verified source data.\n4. **Small, reproducible fixtures.** Every benchmark should isolate one failure mode and have explicit expected findings.\n5. **Tool-agnostic core.** Codex, Claude Code, Cursor, and other agents should consume the same protocol.\n6. **Safe patches over broad rewrites.** Prefer the smallest change that fixes a verified issue.\n\n## Roadmap\n\n- More WooCommerce and Shopify fixtures\n- JSON-LD graph validation\n- Lighthouse\u002Faxe adapters\n- Playwright storefront fixtures\n- Agent-output scoring and rubric-based evals\n- Pull-request review examples using Codex and other coding agents\n- Community-submitted commerce regression cases\n\n## Contributing\n\nSee [CONTRIBUTING.md](CONTRIBUTING.md). New fixtures should be minimal, deterministic, documented, and free of private merchant data.\n\n## Security\n\nDo not submit API keys, customer data, order exports, private themes, or merchant credentials. See [SECURITY.md](SECURITY.md).\n\n## License\n\nMIT. See [LICENSE](LICENSE).\n","commerce-agent-bench 是一个面向电商代码维护场景的编码智能体（coding agent）评测与验证基准框架。它提供可复现的技能定义、确定性回归测试用例和标准化评估流程，支持 WooCommerce、Shopify 及静态电商站点的 PR 审查、事实安全性校验（如价格、库存、评论等硬编码或虚构声明）及平台特异性回归检测。核心采用便携式技能协议（.agents\u002Fskills\u002F）、轻量扫描器和 GitHub Actions CI 集成，强调证据优先、本地可运行、CI 可集成。适用于 AI 编程助手在电商技术栈（如主题开发、SEO 优化、结构化数据修复、无障碍合规）中的能力验证与持续质量保障。",2,"2026-08-15 02:30:14","CREATED_QUERY"]