Personal collection of agent skills, open to steal from (no quality guarantees)
Skill Claude CodeCodex
Use when analyzing Claude Code session logs to extract insights, debug agent behavior, measure performance, or audit tool usage — provides jq recipes for the JSONL transcript format.
Skill Claude CodeCodex
Use when debugging a running bound agent instance — covers database inspection, MCP-based probing, TDD fix cycles, and techniques for discovering issues through live agent conversation.
Skill Claude CodeCodex
Use this before making any commit as the user has special formatting requirements that need to be followed for consistency across projects. If you ask to commit code you just wrote, or if the user asks you to commit something, you should check this skill first to understand how the user wants you to do it.
Skill Claude CodeCodex
Use when creating a new project, repository, package, or application from scratch - ensures official scaffolding tools are used instead of hand-writing boilerplate.
Skill Claude CodeCodex
ALWAYS use before starting a creative task.
Skill Claude CodeCodex
Use when investigating how external or open-source projects implement features, make design decisions, or evolve over time — guides research strategies across code, issues, and documentation.
Skill Claude CodeCodex
Use when collecting and organizing pull request review feedback - provides GitHub MCP gateway patterns for fetching PR comments, review threads, and issue comments, then consolidating into actionable revision lists.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: