kauffj

find-session

Find a past or active Claude Code session by searching transcript text — accepts a grep string, a regex, or a natural language description of the session

humanizer

Rewrites text to remove the patterns that mark it as AI-generated and adds a real human voice in their place. Use when the user asks to "humanize", "de-slop", "de-AI", "make this sound human", "remove the AI tells", or says a draft "sounds like ChatGPT" and wants it fixed. Reasons over clusters of tells weighted by durability, then rewrites. Do NOT use for general copyediting, grammar fixes, proofreading clean human prose, translation, or summarization. This is AI-tell removal plus voice, not a line-edit service.

kauffj-voice

Write or rewrite text in Jeremy Kauffman's voice. Two registers — punchy (tweets, default) and serious (essays, docs, arguments; intellectual, never crass or mean). Use when generating a tweet, or rewriting copy that "reads too AI" into Jeremy's voice.

DevOps & Infrastructure Listed

fix-ci

Diagnose and fix failing CI checks on the current branch

feature

Build and ship a user-facing feature end-to-end — workspace → plan → implement → automated review → screenshots → manual review → direct delivery. Use when asked to build, add, or implement a feature substantial enough to want a branch and a review pass (not a one-line tweak), or to list, resume, complete, abandon, or remove an in-flight feature. Thin orchestrator over the workspace, propose, and review-pr skills; stops for user approval at the plan and manual review, then ships directly.

propose

Produce an approved implementation proposal — plan, simplicity-review the plan, checkpoint with user. Use before non-trivial implementation work (architectural decisions, multi-file changes, features) when alignment is wanted before writing code. Returns a plan ready to implement.

review-pr

Run specialized review agents (security, simplicity, UI, optional visual + QA) over a PR, a worktree, or an explicit file set.

workspace

Manage isolated parallel work items — each a git worktree + dev-server port + copied env + isolated DB + state record. Use when starting work that should live off the main checkout (a feature, bug fix, refactor, or spike on its own branch), when resuming or listing in-flight work items, or when finishing/abandoning/removing one and tearing down its resources. Also the primitive layer the feature, propose, and review-pr skills use instead of reimplementing worktree, port, or DB setup. Subcommands — new, list, get, resume, done, abandon, remove.

explore

Deep codebase research in isolated context — investigate a question thoroughly without cluttering the main conversation

review-security

Review changed code for authorization gaps, input validation failures, data exposure, and injection vulnerabilities

review-simplicity

Review changed code for unnecessary complexity, complection, premature abstraction, and incidental complexity

Web & Frontend Listed

review-ui

Review changed frontend code against UI design principles (Nielsen heuristics + Atomic Design)

Data & Documents Listed

writing

Jeremy's longform writing process, preferences, and human-in-the-loop Google Doc collaboration protocol. Use when helping draft, structure, outline, scaffold, edit, fact-check, or reorganize any essay, article, blog post, Substack piece, tweet thread, or long-form argument Jeremy will publish under his own name — or when collaborating in a Google Doc on such a piece. Also fires on oblique openers like "let's get back to the essay / the AI piece." Jeremy writes the actual prose himself; this skill is scaffolding + evidence + editing his words, NOT ghostwriting. Do NOT use for code comments, commit messages, PR descriptions, docs-as-code / READMEs, or short throwaway/parody tweets (that's kauffj-voice).