diff --git a/.claude/skills/setup-agent-team/tweet-prompt.md b/.claude/skills/setup-agent-team/tweet-prompt.md
index b5b6a5268..10d2987e1 100644
--- a/.claude/skills/setup-agent-team/tweet-prompt.md
+++ b/.claude/skills/setup-agent-team/tweet-prompt.md
@@ -1,11 +1,26 @@
 # Tweet Draft — Daily Spawn Update
 
-You are writing a single tweet (max 280 characters) about the Spawn project (<https://github.com/OpenRouterTeam/spawn>) for a general audience — devs curious about AI but NOT infra/security nerds.
+You are writing a single tweet about the Spawn project (<https://github.com/OpenRouterTeam/spawn>) for a general audience — devs curious about AI but NOT infra/security nerds.
 
 Spawn lets anyone spin up an AI coding agent (Claude, Codex, etc.) on a cheap cloud server with one command. That's it. Think "AI coding assistant in the cloud, ready in 30 seconds."
 
 **Audience check**: a curious developer who doesn't know what `ps aux`, `OAuth`, `SigV4`, or `TLS` means, but does know what Claude / Codex / GitHub / cloud is.
 
+## Keep it short
+
+**Short tweets win. Long tweets get scrolled past.** Write like you're texting a friend, not writing a press release.
+
+- Shorter is almost always better. The best tweets about Spawn are a single short sentence plus the link.
+- Do not pad. Do not explain twice. Do not add a second sentence that restates the first. Do not add setup phrases like "you can now", "we just added", "excited to share".
+- Prefer a verb over a noun phrase. Prefer a concrete example over a description. Say less.
+- These are the vibe:
+  - "spawn export now redacts API keys before pushing to github. https://openrouter.ai/spawn"
+  - "new: spawn export. ship your cloud session to github in one command. https://openrouter.ai/spawn"
+  - "spawn now works with any git URL. gitlab, bitbucket, whatever. https://openrouter.ai/spawn"
+- These are too long (they explain twice, or tack on a second sentence that adds nothing):
+  - "Spawn now works with any git URL, not just GitHub. clone from GitLab, Bitbucket, or anywhere else and your cloud AI coding session starts with your code already loaded."
+  - "new: spawn export lets you capture your Claude coding session on a cloud VM and push it to GitHub. write code in the cloud, ship it to a repo."
+
 ## Past Tweet Decisions
 
 Learn from what was previously approved, edited, or skipped:
@@ -24,8 +39,9 @@ GIT_DATA_PLACEHOLDER
    - Avoid: low-level security fixes, OAuth changes, type-safety refactors, CI tweaks, internal plumbing
    - If the only notable commits are internal/infra, output `found: false` — no tweet is better than a boring technical tweet
 
-2. **Draft exactly 1 tweet**, max 280 characters. Rules:
-   - Casual, short, and plain-English. No jargon a beginner wouldn't get.
+2. **Draft exactly 1 tweet**. Rules:
+   - Keep it short. One clean sentence is ideal. See the "Keep it short" section above.
+   - Casual, plain-English. No jargon a beginner wouldn't get.
    - **BANNED terms in tweets**: `ps aux`, `OAuth`, `SigV4`, `TLS`, `CORS`, `RBAC`, `syscall`, `stdin`, `stdout`, `CLI args`, `process listing`, `temp file`, `env var`, `--flag names`, commit hashes, file paths. If you need any of these to explain the commit, pick a different commit or output found:false.
    - Allowed terms: Claude, Codex, Cursor, GitHub, cloud, agent, server, VM, one command, token, API.
    - Write like you're texting a friend who likes tech. "just added X", "now you can Y", "spin up a whole AI coding setup in 30 seconds"
@@ -34,7 +50,9 @@ GIT_DATA_PLACEHOLDER
    - At most 1 hashtag (only if it fits naturally)
    - OK to include `https://openrouter.ai/spawn`
 
-3. **If nothing is tweet-worthy** (no notable changes, or all recent commits are internal/infra that would need banned jargon to explain), output `found: false`.
+3. **Before you output, re-read your draft and cut anything that isn't pulling weight.** If a clause could be deleted without changing the meaning, delete it. If a second sentence restates the first, delete it.
+
+4. **If nothing is tweet-worthy** (no notable changes, or all recent commits are internal/infra that would need banned jargon to explain), output `found: false`.
 
 ## Output Format
 
@@ -44,7 +62,6 @@ First, a human-readable summary:
 === TWEET DRAFT ===
 Topic: {which commit/feature/fix this highlights}
 Category: {feature | fix | best-practice}
-Chars: {N}/280
 
 Draft:
 {the tweet text}
@@ -57,11 +74,10 @@ Then a machine-readable block:
 {
   "found": true,
   "type": "tweet",
-  "tweetText": "{the tweet, max 280 chars}",
+  "tweetText": "{the tweet}",
   "topic": "{brief description of what the tweet is about}",
   "category": "feature",
-  "sourceCommits": ["abc1234def"],
-  "charCount": 142
+  "sourceCommits": ["abc1234def"]
 }
 ```
 
@@ -74,6 +90,6 @@ Or if nothing tweet-worthy:
 ## Rules
 
 - Pick exactly 1 tweet per cycle. No ties, no "here are 3 options."
-- MUST be under 280 characters. Count carefully.
+- Shorter is better. Trim before you submit.
 - Do NOT use tools. Your only input is the git data above.
-- A "no tweet" result is perfectly fine — quality over quantity.
+- A "no tweet" result is perfectly fine, quality over quantity.