moiri-gamboni

humanizer

Remove signs of AI-generated writing from text. Use when editing or reviewing text to make it sound more natural and human-written. Based on Wikipedia's comprehensive "Signs of AI writing" guide. Detects and fixes patterns including: inflated symbolism, promotional language, superficial -ing analyses, vague attributions, em dash overuse, rule of three, AI vocabulary words, passive voice, negative parallelisms, and filler phrases.

API & Backend Listed

havelock-api

This skill should be used when writing prose for an external audience and aiming for a warm, conversational, or human tone rather than stiff corporate language. Also use when the user explicitly asks to "analyze orality", "check orality score", "use Havelock", "use the Havelock API", or mentions "havelock.ai", "orality detection", or "oral vs literate". Havelock scores text on a literate-to-oral spectrum, useful for calibrating tone toward natural, approachable writing.

html2markdown

This skill should be used when the user asks to "convert HTML to markdown", "scrape a webpage to markdown", "download a page", "download a webpage", "extract content from HTML", "pipe HTML to markdown", "batch convert HTML files", or mentions "html2markdown", "html-to-markdown", or converting/downloading websites/pages to markdown format. Do NOT trigger on "fetch" (that implies WebFetch, not CLI conversion).

pastebin

Publish text or markdown as an end-to-end-encrypted paste with a public shareable link, using the self-hosted PrivateBin at https://paste.${DOMAIN}/. Use when the user asks to publish, share, or paste a document, note, report, or snippet for someone else — especially recipients outside the tailnet. Also covers reading and deleting existing pastes via pbincli.

AI & Automation Listed

usage-limits

Check this Claude plan's live usage limits — the 5-hour and weekly (7-day) rate-limit %s and reset times that actually gate the session, plus context-window fill. Has a `--guard` pacing gate (per-window configurable) for multi-agent fan-outs. Use mid-task to decide whether to keep spawning subagents, before/between waves of a big job, or when the user asks "how close are we to the limit", "how much is left", "when does it reset", or "how full is the context". A compact one-liner (date/time + 5h & weekly %s) is wired as a per-turn hook (`usage --hook`, transcript-suppressed), so every prompt already carries the current time and live usage. Runs the `usage` command (alias `roost-usage`).

AI & Automation Listed

codex

Delegate a task to OpenAI Codex (GPT) as a headless subagent — `codex exec` does the work on the user's OpenAI/ChatGPT credits instead of Claude's rate-limit budget. Use when the user says "ask codex", "ask gpt", "delegate to codex/gpt", "have gpt do it", wants a second-model opinion or review, or when a Claude usage cap is near/at 100% and work should be offloaded. Covers one-shot and background runs, follow-ups (resume), sandbox levels, parallel fan-out, code review, and auth.

DevOps & Infrastructure Listed

session

The session CLI — this Claude Code session's identity and its usage of the plan's rate limits. Run `session` for the overview (id · name, the 5-hour and weekly rate-limit %s + reset times that actually gate the session, context fill, this session's share); `session whoami` for the id/title (pane-safe, e.g. to build a `claude --resume <id>` command or label output); `session usage [--all]` for per-session burn attribution; `session --guard` as a pacing gate for multi-agent fan-outs. Use when the user asks "how close are we to the limit", "how much is left", "when does it reset", "how full is the context", "how much has this/each session used", "which session is eating the budget", or "what's my session id / name" — and mid-task before/between waves of a big job to decide whether to keep spawning subagents. A per-turn hook (`session --hook`, transcript-suppressed) stays silent below `USAGE_WARN_PCT` (default 90) and injects the usage line + a ⚠ advisory only once a window crosses it — so absent a ⚠, usage is un

roughdraft

Put a Markdown file in front of the user for review in a browser, then read their comments and suggested edits back and reply inside the document. Use when the user wants to review, comment on, or mark up a document, when handing over a plan or draft for feedback, or when they say "open this in roughdraft" or "rd". Covers opening the document, waiting for Done Reviewing without losing the signal, and the CriticMarkup shapes that actually render.

zotero

This skill should be used when working with a Zotero library — summarizing or reading attached PDFs, organizing items into collections, tagging, adding notes, linking related items, or batch-processing references. Trigger on "Zotero", "my library", "references", "organize my papers/notes", "summarize these PDFs" in a research-library context. Covers the local read API, pyzotero for writes, and direct storage reads for bulk PDF work.