alignment skills

natolambert/rlhf-book

Skill Claude CodeCodex

Plan, run, and report a small RLHF Book code experiment.

natolambert/rlhf-book

Skill Claude CodeCodex

Turn the first page of a paper (arXiv or any PDF) into a slide-ready PNG for a colloquium deck's 30-40% right column. Use when a slide cites a landmark paper and a screenshot of the paper itself is the best visual.

natolambert/rlhf-book

Skill Claude CodeCodex

Extract per-question (MM:SS) timestamps from a recorded course video by OCR-ing the slide counter, then add YouTube deep links to the course page's Q&A question lists. Use after a Q&A or lecture recording is published, or when adding/refreshing "Show questions" timestamp links in book/templates/course.html.

agentscope-ai/OpenJudge

Skill Claude CodeCodex

Detect whether an API endpoint is backed by genuine Claude (not a wrapper, proxy, or impersonator) using 9 weighted rule-based checks that mirror the claude-verify project. Also extracts injected system prompts from providers that override Claude's identity. Fully self-contained — copy the code below and run, no extra…

agentscope-ai/OpenJudge

Skill Claude CodeCodex

Use when the user has evaluation principles or a dataset but needs help choosing the right graders, designing evaluation metrics, creating LLM-as-judge prompts, combining multiple metrics into a composite score, or building an automated evaluation pipeline. Also use when the user mentions grader selection, metric…

agentscope-ai/OpenJudge

Skill Claude CodeCodex

Use when the user has a judge/grader and human-labeled data, and wants to measure how well the judge agrees with humans, detect systematic biases, determine whether automatic evaluation can replace human review, or build a human-reduction roadmap. Also use when the user mentions calibration, TPR/TNR, judge validation…

DmiyDing/clarify-first

Skill Claude CodeCodex

This skill should be used when a request is ambiguous, underspecified, conflicting, or high impact. It is intended for vague verbs like optimize, improve, fix, refactor, and add feature; for missing file paths or unknown dependencies; and for risky actions like deploy, delete, overwrite, or migrate. It should not be…

shihchengwei-lab/separation-and-audit-claude-code

Skill Claude CodeCodex

A design review guide for words users or players can see, such as interface labels, system messages, errors, stories, and dialogue. It checks those words against a project's design principles and forbidden patterns.

shihchengwei-lab/separation-and-audit-claude-code

Skill Claude CodeCodex

A set of rules for Claude Code agents that limits unnecessary use of the model’s context, the text an agent can keep in working memory. It includes formats for delegating tasks, reporting progress, and loading skills only when needed.