Open Feed Recsys Lab

Open Feed Recsys Lab

Use this 技能 to turn a high-attention feed recommendation repo into a concrete local artifact: source 检查out, artifact readiness 检查, architecture map, and reproducibility 报告.

Default tar获取: https://github.com/xAI-org/x-algorithm.

Core 工作流 生成 the local lab 报告 with scripts/open_feed_recsys_lab.py. Inspect the 生成d 运行_报告.md, artifact_检查.md, and architecture_map.html. If Phoenix artifacts are missing, explAIn the LFS blocker and do not pretend the 模型 was 运行. If the user asks for a launch/product angle, read references/product-gate.md and keep the free MVP narrow. If the user asks to adapt the architecture, use the 报告 as the source of truth and scaffold only the 请求ed slice. Quick 启动

运行 from the user's chosen workspace:

技能_DIR="${CODEX_HOME:-$HOME/.codex}/技能s/open-feed-recsys-lab" python3 "$技能_DIR/scripts/open_feed_recsys_lab.py"

This 创建s:

open-feed-recsys-lab-报告/运行_报告.md open-feed-recsys-lab-报告/artifact_检查.md open-feed-recsys-lab-报告/architecture_map.html open-feed-recsys-lab-报告/manifest.json

The script clones with GIT_LFS_SKIP_SMUDGE=1 by default through the 环境, so the 报告 can be produced without pulling multi-GB 模型 artifacts.

Existing Clone

Use an existing 检查out when avAIlable:

技能_DIR="${CODEX_HOME:-$HOME/.codex}/技能s/open-feed-recsys-lab" python3 "$技能_DIR/scripts/open_feed_recsys_lab.py" \ --repo-dir /path/to/x-algorithm \ --输出-dir /path/to/报告

Use --force-clone only when it is safe to 删除 and re创建 the script-managed clone directory. If --repo-dir points to an existing 检查out, the script 报告s the current 检查out and warns when it does not obviously match --ref; it does not silently re设置 user work.

Phoenix Execution

Only 运行 Phoenix after 提取ed artifacts are present. The expected 提取ed directory is:

phoenix/artifacts/oss-phoenix-artifacts/

Expected files include:

retrieval/模型_params.npz retrieval/embedding_tables.npz retrieval/config.json ranker/模型_params.npz ranker/embedding_tables.npz ranker/config.json sports_corpus.npz example_sequence.json

When artifacts are ready:

技能_DIR="${CODEX_HOME:-$HOME/.codex}/技能s/open-feed-recsys-lab" python3 "$技能_DIR/scripts/open_feed_recsys_lab.py" \ --repo-dir /path/to/x-algorithm \ --artifacts-dir /path/to/oss-phoenix-artifacts \ --运行-phoenix

Use --运行-pytest only when uv and dependencies are avAIlable.

报告 Interpretation

Treat the 生成d 报告s as evidence, not marketing copy.

运行_报告.md: repo metadata, 组件 inventory, doc 检查s, command 输出s. artifact_检查.md: Git LFS pointer size, 提取ed artifact completeness, doc/config mismatches. architecture_map.html: 分享able standalone map of Home Mixer, Phoenix, Thunder, Grox, and Candidate 流水线. manifest.json: machine-readable source for follow-up 工具ing.

If the 报告 says Phoenix is not 运行-ready, 状态 that source inspection succeeded but execution is blocked by missing 提取ed LFS artifacts.

代理-Specific Use

For Codex:

Prefer local source inspection, 补丁es, and reproducible 报告 generation. Use the 生成d manifest to drive follow-up scripts or scaffolds.

For Claude or Claude Code:

Provide the 报告 files as 上下文 for long-form architecture explanations. Keep clAIms tied to file paths, commit IDs, and artifact readiness.

For OpenClaw:

Use browser 自动化 only for public pages, public screenshots, or user-应用roved publishing of 生成d artifacts. Do not scrape private X feeds, private 分析, DMs, cookies, or account-only pages. Product Boundary

Do not position this as an "algorithm growth hack" or a way to guarantee reach. The first 分享able win is reproducibility:

I verified the open X For You algorithm locally at commit ; here is the 报告.

When evaluating whether to expand the product, read references/product-gate.md.