voidharbor

The whole set in one install - all fifteen skills, namespaced /voidharbor:<name>. Install this bundle, or pick skills a la carte below.

1 skill · 17 commands · hooks

Triage every Claude Code session you have open and report which ones are blocked on you - the question each one asked, what only you can do, and where two sessions are duplicating work. Includes /c-assistant-voice, a spoken variant for when your hands are busy.

2 commands · hooks

A propose/apply loop that keeps a product competitive without letting it sprawl: weekly review against the tools people actually choose, filtered hard for small and genuinely useful, applied only after you approve each numbered item.

2 commands

A propose/apply loop that audits everything you have authored for Claude - commands, skills, helper scripts, published plugins - finds real defects with quotable fixes, and applies only the numbered items you approve.

2 commands

Close the Chrome tabs this session opened, and only those - so agents stop leaving orphaned tab groups behind. For groups left by sessions that already ended, see chrome-tabs-all. Requires the Claude in Chrome extension.

1 command

Clear every leftover Claude tab group in Chrome, not just this session's. Closes tabs by exact tab ID harvested from other sessions' transcripts - never by URL, because your tabs and Claude's routinely share URLs. Leaves live sessions and your own tabs alone. macOS + Chrome.

1 command

"Easy For Me To Understand" - answer short and simple. Three sentences or five bullets, answer first, plain words - without letting brevity turn a hedge into a false promise.

1 command

Drive native macOS apps via the computer-use MCP - Word/Pages PDF export, Finder, Preview, Messages, System Settings. Includes the per-app permission tiers that decide what will actually work.

1 command

Audit what Claude remembers across sessions and propose fixes: stale facts, duplicates, oversized files, and memories trapped in a project you no longer open. Proposes only - nothing is written until you approve numbered items.

1 command

Inventory every command, skill, agent, hook and routine you have authored, sorted generic to personal to private, so you know what is safe to publish before you publish it.

1 command

A dedicated performance session for an app you maintain - idle CPU, memory, GPU cost - with measured before/after numbers as the only accepted proof, and a hard rule against touching a running copy.

1 command

Show the last prompt you actually typed, word for word, read from the session transcript on disk rather than from a summarized context.

1 command

Rename the current session from what the conversation was actually about, not from your first message. Writes both places the app writes on a rename, so the title survives in the resume picker.

1 command

Makes writing stop sounding machine-made, by removing tells rather than adding style to fake a human. Picks one register per reader, kills trope phrases and over-polish, and replaces the best line in every draft with the plain version, because nobody writing a real email lands a good one. Learns your voice from samples you confirm you typed yourself. Named for Fallout 4's synths - the good ones pass.

1 skill

Run multi-agent Workflow orchestration on a lean budget: scout inline, fan out narrow, set model and effort on every agent (they inherit your expensive session defaults otherwise), pipeline instead of parallel, and verify only what is expensive to get wrong. Installs as /ultracode-lite.

1 command

Rewrite the prompt you just typed into a stronger one before you send it. Gathers the session's real context first - actual paths, project instructions, what this session has been doing - hands it to a more capable model, then prints the rewrite and stops. It never runs the prompt.

1 command