lede

Turn raw notes and news into an editorial, emoji-accented message formatted for Discord and Telegram, ready to copy-paste and send yourself as a user (no bot, no API). Prose is linted against AI-slop with vale.

It's an agent skill in the standard SKILL.md format, so it runs in Claude Code, Hermes, or any harness that loads skills: install it, then ask the agent to turn a pile of notes into two send-ready channel posts.

What it produces

Feed it raw input (notes, bullets, links, news). It writes one clean editorial fact brief, then independently writes and lints two platform-specific messages:

• Discord: native markdown (## headers, - bullets, [masked](links)). Renders when you paste and send.
• Telegram: native bold + emoji + - bullets + bare URLs. The Telegram rendering surface creates normal formatting entities that survive this workflow's copy-paste. Unicode mathematical bold is forbidden.

You copy each message and send it into the channel yourself. Discord is shown in a fenced block so its markdown stays literal. Telegram is shown as ordinary rendered text, never as a code block, so native formatting is preserved.

Requirements

Need	Why	Required?
An agent that loads SKILL.md skills	Claude Code, Hermes, etc.	yes
git	to clone / update the skill	yes (install only)
Python 3.7+	scripts/lede.py: formatting validation + message length checks	recommended
vale	the AI-slop lint gate	recommended

The skill still runs with neither Python nor vale, but loses automated formatting, length, and prose checks. Install both for the full experience.

Install

Claude Code

Clone straight into your skills folder, then start a new session:

# macOS / Linux
git clone https://github.com/sasonov/lede.git ~/.claude/skills/lede

# Windows (PowerShell)
git clone https://github.com/sasonov/lede.git "$env:USERPROFILE\.claude\skills\lede"

Update later: git -C ~/.claude/skills/lede pull

Hermes

Skills live under ~/.hermes/skills/<category>/<name>/. Clone into the communication category (matches the skill's own metadata.hermes.category):

git clone https://github.com/sasonov/lede.git ~/.hermes/skills/communication/lede

Or use the CLI, which resolves GitHub skill repos and runs a safety scan:

hermes skills install sasonov/lede --category communication

Then hermes skills list should show lede. Update later with git pull in that folder.

Install the dependencies

vale: the anti-slop gate (no vale sync needed; the styles ship in the repo):

OS	Command
macOS / Linux	brew install vale
Windows	winget install errata-ai.Vale or scoop install vale
any	download a release

Python 3.7+: from your package manager or python.org. Most systems already have it.

Verify

# adjust the path to wherever you installed the skill
python ~/.claude/skills/lede/scripts/lede.py --selftest   # prints: ok
vale --version

Use

In Claude Code, just describe the task: the skill triggers on its own:

Turn these release notes into a Discord and Telegram post: (paste your notes)

…or invoke it by name: "use the lede skill on this." Optional tone flag: calm (default) or punchy. Claude drafts, lints, and hands you the two send-ready blocks. Copy → paste → send.

Example

Input

- v2.3 shipped today
- new dark mode, 40% faster cold start
- breaking: old API keys stop working aug 1
- docs: example.com/v2

Discord

## 🚀 v2.3 is live

**Dark mode** landed, and cold starts are **40% faster**.

**Heads up, breaking change:** old API keys stop working **August 1**. Rotate
yours before then.

Docs: [example.com/v2](https://example.com/v2)

Telegram

⚠️ Rotate old API keys by August 1

Old API keys stop working on August 1. Rotate yours before then.

🚀 v2.3 shipped today with dark mode and 40% faster cold starts.

📚 Docs: https://example.com/v2

How it works

1. Extract one platform-neutral fact brief: facts only, not reusable prose.
2. Author Discord and Telegram independently with different information architecture.
3. Review both drafts against the mandatory structural anti-slop rubric.
4. Lint both drafts separately with Vale against AI-slop wordlists.
5. Validate structure, formatting, and length with scripts/lede.py check, then enforce separate authorship with scripts/lede.py compare.

Message text is always passed to the helper via --file (or stdin), never as a shell argument: so pasted content with $(...), backticks, or | can't be executed by the shell.

The scripts/lede.py helper

python scripts/lede.py check discord  --file discord.txt
python scripts/lede.py check telegram --file telegram.txt
python scripts/lede.py compare --discord-file discord.txt --telegram-file telegram.txt
python scripts/lede.py count discord  --file discord.txt
python scripts/lede.py count telegram --file telegram.txt
python scripts/lede.py --selftest                       # -> ok

Files

SKILL.md                  the instructions the agent follows
references/formatting.md  native Discord + Telegram formatting rules
scripts/lede.py           formatting validator + per-platform length checker
.vale.ini                 vale config (local styles, no `vale sync`)
styles/Editorial/*.yml    lexical slop, hedging, urgency, and punctuation rules

Telegram accessibility

Use native Telegram bold only. Unicode mathematical alphanumeric characters are not real formatting, harm search and screen-reader output, and are rejected by the validator. The validator also rejects wrapping the entire Telegram message in a code block, blockquote, or indented code block.