Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/jigripokri/poha/draftnpx skills add jigripokri/POHA --skill draftgit clone --depth 1 https://github.com/jigripokri/POHAWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00035 | $0.00519 |
| Opus 5 | $0.00017 | $0.00260 |
| Sonnet 5 | $0.00007 | $0.00104 |
| Haiku 4.5 | $0.00003 | $0.00052 |
Grade A, and why
draft scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
What it actually says
/draft
Write outbound messages in the user's voice. Used on demand for any thread.
Recipe
When invoked:
-
Identify the thread. User says "draft a reply to " or "draft a message to about ." If ambiguous, ask one clarifying question.
-
Read the last 10–20 messages in the thread. Use the appropriate MCP (WhatsApp, Gmail, Android Messages).
-
Study voice patterns.
- Sentence length and structure
- Vocabulary level (formal vs. casual)
- Regional slang or in-group references
- Punctuation habits
- Use of emoji or not
- Greeting/sign-off patterns (or lack thereof)
-
Read CLAUDE.md > Voice & Drafting Authorization.
-
Read memory/people.md for context on the recipient (if available).
-
NEVER read memory/private/carry-on.md. Do not reference anything from it.
-
Draft 1–3 options. If user said "draft" (singular), give one. If user said "options" or the message is high-stakes, give 2–3 with brief notes on the difference.
-
Show the draft(s). Do NOT send.
-
Wait for approval. "Send it", "send A", "send the second one" = green light. Anything else means iterate.
-
On approval, send via the appropriate MCP.
Voice rules (default — override in CLAUDE.md)
- Direct. No "Hope this finds you well."
- Match the recipient's tone — if they're casual, you're casual.
- Reference something specific from the recent conversation (callback, shared joke, thing they mentioned).
- Short. Two paragraphs maximum unless the topic genuinely requires more.
- No AI tells: "As an AI...", "I hope this helps...", "Let me know if you have any questions!" — never.
What this skill won't do
- Send before "send it" approval. Always wait.
- Quote anything from carry-on.md.
- Send to multiple recipients without confirming the list explicitly.
- Send if the draft contradicts something obvious from memory (e.g., agreeing to a date the user has a conflict on).
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 54 lines · 35 tokens per session scan A fc0f5520e9fb
draft is a skill published in the GitHub repository jigripokri/POHA (118 stars, last pushed 3mo ago), licensed MIT. It adds 35 tokens to every session and 519 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
rich-post
Send a Telegram rich-media post to ANOTHER allowlisted chat via the bot (Bot API 10.1 sendRichMessage) - text, inline images, tables, headings, lists, quotes, collapsible blocks, formulas, collages/slideshows ALL in one message bubble. Use when the owner asks to post somewhere other than the current conversation (the…
telegram-userbot
Подключение и работа с личным Telegram-аккаунтом владельца (userbot, НЕ бот). Используй, когда просят читать личные чаты/переписку/диалоги, искать по своим сообщениям, писать кому-то от своего имени, «зайди в мой телеграм», «подключи мой телеграм», «прочитай личку», «ответь в личке», «перешли сообщение», а также при…
memory-processor
Iva's daily-memory processor. Reads the day's two-sided transcript (daily/YYYY-MM-DD.md), distills noteworthy entities / decisions / ideas into typed autograph cards, links them into the graph, and produces a daily-summary card (topics + MOC) that navigates down to the raw transcript and up to the week. Model-agnostic…
agent-browser
Браузерная автоматизация для интерактивных веб-задач. Используй, когда нужно открыть сайт, заполнить форму, кликнуть, сделать скриншот, залогиниться, спарсить JS-страницу или протестировать веб-приложение. Триггеры — «зайди на сайт», «заполни форму», «нажми кнопку», «скриншот страницы», «войди в аккаунт», «спарси…
update-recovery
Restore or merge user customizations preserved during an Iva update. Use when the owner says "restore my update changes", "recover my customizations", «верни мои изменения после обновления», «восстанови мои доработки» or asks about an update conflict.
security-defense
Procedures for handling untrusted content, suspected prompt injection, secret exposure, data exfiltration, and manual security inspection. Use for emails, web pages, forwarded messages, webhooks, group content, or a suspected leak. Do not use as a general claim that every command or model call is automatically…