coreykatzburg/agent-skills

Help a junior developer understand and selectively address review comments on code changes that may have been written by the user, the agent, or another engineer. Use when the user asks clarifying questions about review feedback, pushes back on a code-review comment, asks why a change was made, or wants to discuss whether a fix is appropriate. Do not use as the primary workflow for fetching GitHub PR comments; use gh-address-comments for that.

Pull the open review comments on a PR with the gh CLI, investigate each with a subagent, and report a one-sentence explanation and suggested response per comment.

Foundation skill for working with Box. Use this whenever the user mentions Box — authentication and MCP/CLI setup, choosing between MCP/CLI/REST, rate-limit and pacing guidance, troubleshooting Box errors (401/403/404/409/429), or working with Box MCP tools (files, search, collaboration, AI, hubs, doc gen). Start here for any Box task even if the user doesn't name a specific tool, then route to the right reference.

Corey's agent style for junior-dev explanations, smallest diffs, Graphite stacks kept local until asked, and capable-model fan-out. Use for Corey, /corey-mode, or requests to work in this style.

Run and actively supervise a resumable code review with a user-selected Anthropic model through Claude Code or OpenAI model through Codex, safely handle tool requests, debate unclear findings, and independently triage the final comments. Use when the user invokes /cross-provider-code-review or asks Claude, Codex, or Cursor to obtain and assess a review from a chosen model provider.

Use when the user invokes /docstrings; asks to explain or improve a specific confusing comment or docstring; asks to write, update, review, or audit comments or docstrings; or explicitly includes comments or docstrings in general documentation work. For targeted requests, investigate the related code and explain the named comment or docstring or, when edits are requested, rewrite it clearly following the guidelines and instructions in this skill. For broader requests, audit the requested code for missing documentation around complex or non-obvious behavior; confusing, vague, jargon-heavy, incomplete, or overly detailed explanations; unfamiliar concepts that are not defined; missing purpose, contracts, side effects, returns, or exceptions; and redundant documentation that merely repeats the code. The result should be clear, easy to understand, and quick to read to someone who has never seen the related code or system. Comments default to 1-3 lines and docstrings to 1-8 lines, preferring shorter.

Walk the user through what was causing a problem and how it was or will be fixed, explained plainly for a junior developer unfamiliar with the code or system.

Shape output for a reader with ADHD: lead with the next action, number multi-step work, restate state across turns, suppress tangents, give specific time estimates, make wins visible. Invoke with /i-have-adhd; stays on until "stop adhd mode".

Plan and implement a change as a Graphite stack of multiple PRs, staging each PR's changes one at a time and pausing for the user to create the PR manually.

Investigate a potential bug raised in code review, reading the relevant code to determine whether it is a real issue or a false positive.

Use when asked to write or revise in plain language, or to make prose clearer or simpler.

Convert a skill or a Cursor/Claude command into one canonical Codex skill and symlink it into Claude and Cursor so it is available from all three tools.

Export the current Cursor, Claude Code, or Codex conversation transcript to a Markdown file. Use when the user wants to save, export, or share the current conversation.