Max-Levitskiy

analyze-density

Analyze text for information density, semantic repetition, and filler content. Two modes: "score" (read-only audit with per-method breakdown) and "fix" (rewrite to remove redundancy while preserving all unique information). Dispatches 7 core analysis methods (plus genre-targeted ones) as parallel subagents, then aggregates results into a composite score and actionable report. Use when asked to "check density", "find repetition", "remove AI slop", "deduplicate this text", "is this repetitive", "information density", "compress this writing", "tighten this doc", or whenever text feels bloated or AI-generated. Also use proactively when reviewing docs or PRs that contain suspiciously fluffy prose.

0 Updated 1 months ago

analyze-density

0 Updated 1 months ago

analyze-code-density

Analyze code for AI-generation slop: duplication, dead code, redundant comments, verbosity inflation, over-abstraction, error masking, convention violations, hallucinated dependencies, and performance waste. Score mode only: dispatches 10 analysis methods as parallel subagents and aggregates a composite 0-100 density score with per-method findings. Works on a git diff/branch/PR with repo context, or on standalone files. Use when asked to "check code density", "find AI slop in this code", "is this code bloated", "review this AI-generated code", "code slop check", or before merging agent-authored changes.

fellow

Retrieve and work with Fellow (fellow.app) meeting notes, transcripts, AI summaries, and action items — answer questions about past meetings, write recaps into a repo, bulk-export to markdown. Trigger even when Fellow is never named — any request to recall, search, summarize, or file what happened in a meeting, like "what did we decide about X", "my action items this week", "pull Friday's transcript". Also for setting up Fellow API access.

orchestrate

Orchestrate subagents over a parent task as small, tracked, parallel work packages, with an async question protocol so waiting on human decisions never blocks progress. Use whenever the user asks to "orchestrate subagents/agents" on a task or project, wants work done "mostly autonomously" while they step away, says "split it into subtasks and work in parallel", "collect questions for me", "ask me when you need a decision but don't block", or hands over any multi-part deliverable (documents, prep materials, analyses, migrations) that benefits from parallel agents plus occasional human input. Also use when RESUMING such an orchestration: an agent-completion notification arrives, the user answers an open question, or a pending wave needs launching. Trigger even if the user never says "orchestrate" — "work through this task with agents while I'm out" or "keep going async, I'll answer questions later" is this skill.

agent-config

Set up, repair, or relocate the settings and credentials a skill, subagent, command, or hook depends on — layered `.agents/config` files plus credential *references* (1Password, environment variable, dotenv, macOS Keychain, any shell command) that keep secrets out of git. Use when a component reports missing or incomplete configuration, when its credential won't resolve, when the user asks where an API key or setting is stored or wants to move one between the global, repo, and local layers, and when a skill author is adding configuration to a new skill or subagent. Other skills invoke this instead of restating any of it.

herdr

Run and manage AI subagents in herdr panes, and drive the herdr terminal workspace manager from the CLI. Use to start a subagent (claude, codex, omp, gemini, opencode, droid, copilot, cursor - or a preset of your own), send it a task, read its answer, hold a multi-turn conversation with it, list which agents are running and what state they are in, and stop one - one command per step, with the send/Enter and status-timing traps handled. Subagent definitions (argv, cwd, env, session, timeouts) live in agent config, so "run the reviewer agent" resolves to a preset rather than a guessed command line. Also covers the layout itself: list, focus, rename, create, or close workspaces, worktrees, tabs, and panes, and answer "which one am I in". Use whenever the user mentions herdr, asks to run / spawn / ask / message / check on / stop an agent or subagent, or asks to rename / switch / focus / close / list a session, workspace, tab, pane, or worktree - e.g. "run a subagent to review this", "ask the other claude what it