\r \r

Felo Twitter Writer Skill\r

\r

Constraints (MUST READ FIRST)\r

\r These rules are mandatory. Violating any of them will produce incorrect behavior.\r \r

1. This skill uses SuperAgent directly. All generation is handled by felo-superAgent/scripts/run_superagent.mjs with --skill-id twitter-writer. Do NOT attempt to generate tweet content yourself.\r \r
2. ALWAYS use --json flag when calling SuperAgent. In Claude Code's Bash tool, stdout is always captured — it never streams directly to the user. JSON mode returns the full answer in a structured response. After the script finishes, read data.answer from the JSON output and print it verbatim as your response text.\r \r
3. ALWAYS output data.answer verbatim. After the script finishes, print data.answer exactly as-is as your response text. Do NOT summarize, paraphrase, or add commentary around it.\r \r
4. --live-doc-id is REQUIRED for every SuperAgent call. Follow these rules strictly:\r
   • Reuse any live_doc_id already available in this session (from a prior SuperAgent or livedoc call)\r
   • If none: run node felo-livedoc/scripts/run_livedoc.mjs list --json, then find the first item where is_shared === false in data.items (list is sorted by modification time descending, so this gives the most recently modified private LiveDoc). Use its short_id.\r
   • If no is_shared === false item exists (or list is empty): run node felo-livedoc/scripts/run_livedoc.mjs create --name "Twitter Writer" --json, use data.short_id\r
   • NEVER use a LiveDoc where is_shared === true — shared LiveDocs belong to other projects and will cause a 502 error.\r \r
5. Always persist state. After every SuperAgent call, extract thread_short_id and live_doc_short_id from the JSON response fields data.thread_short_id and data.live_doc_short_id. Use them in subsequent calls.\r \r
6. Output language follows the user's input language. Default is en. Detect the user's language and pass the matching --accept-language value: ja for Japanese, en for English, ko for Korean, zh for Chinese. If unsure, use en.\r \r
7. Do NOT pass --timeout to the SuperAgent script. The script manages its own connection lifecycle.\r \r
8. Brand style selection for Mode 2 new conversations only. When starting a new conversation for content creation (Mode 2, no thread_short_id), you MUST attempt to fetch the TWITTER style library and offer the user a style choice BEFORE calling SuperAgent. The style is passed via --ext '{"brand_style_requirement":"..."}'. Full procedure in the Style Selection section below.\r \r
   • Style category is always TWITTER (hardcoded — this skill only writes tweets).\r
   • --ext is only valid for new conversations. Never pass it in follow-up mode (--thread-id).\r
   • If the style library returns no entries, skip silently and proceed without --ext.\r
   • Mode 1 (style DNA extraction) does NOT use this step.\r \r

When to Use\r

\r Trigger this skill when the user wants to:\r \r

• Analyze a Twitter/X account's writing style\r
• Extract a writing style DNA document from tweets\r
• Write, draft, or compose tweets / X posts\r
• Write a Twitter thread (multi-tweet series)\r
• Write an X long-form article / long post\r
• Imitate or mimic someone's tweet style\r
• Ghostwrite tweets on behalf of someone\r
• Understand how a specific account writes\r \r Trigger keywords:\r \r
• English: analyze twitter style, twitter style analysis, extract writing style, style DNA, write a tweet, write tweets, draft a tweet, write a thread, twitter thread, X article, X long post, imitate tweet style, mimic tweet style, tweet in the style of, write like [account], X account analysis, analyze X account, ghostwrite tweets, how does [account] write\r
• 日本語: ツイートを書く, ツイートスタイル分析, スタイルDNA, ツイートを模倣, Xアカウント分析, ツイートのスタイルを抽出, 〇〇風のツイートを書く, ツイートを代筆, Xアカウントを分析, このアカウントはどう書いている\r \r Explicit commands: /felo-twitter-writer, use felo twitter writer\r \r Do NOT use for:\r \r
• General Twitter/X search only (use felo-x-search)\r
• General SuperAgent conversation (use felo-superAgent)\r
• Web search (use felo-search)\r \r

Two Modes\r

\r

Mode 1 — Style DNA Extraction\r

\r When: User provides a Twitter/X account name and wants to understand or extract its writing style.\r \r Steps:\r \r

Step 1: Fetch tweets via felo-x-search\r

\r

node felo-x-search/scripts/run_x_search.mjs --id "USERNAME" --user --tweets --limit 30\r
```\r
\r
Also fetch the account profile:\r
\r
```bash\r
node felo-x-search/scripts/run_x_search.mjs --id "USERNAME" --user\r
```\r
\r
#### Step 2: Obtain live_doc_id\r
\r
Follow Constraint #4 above.\r
\r
#### Step 3: Call SuperAgent with tweet content\r
\r
Determine conversation mode first:\r
- If **no** `thread_short_id` exists in this session → new conversation (pass `--skill-id twitter-writer`)\r
- If `thread_short_id` **already exists** in this session → follow-up (pass `--thread-id`)\r
\r
Replace `LANG` with the user's language: `en` (English), `zh` (Chinese), `ja` (Japanese), `ko` (Korean). See Constraint #6.\r
\r
**New conversation (first call in session):**\r
\r
```bash\r
node felo-superAgent/scripts/run_superagent.mjs \\r
  --query "/twitter-writer ENRICHED_QUERY_WITH_TWEET_CONTENT" \\r
  --live-doc-id "LIVE_DOC_ID" \\r
  --skill-id twitter-writer \\r
  --accept-language LANG \\r
  --json\r
```\r
\r
**Follow-up (thread_short_id already exists):**\r
\r
```bash\r
node felo-superAgent/scripts/run_superagent.mjs \\r
  --query "/twitter-writer ENRICHED_QUERY_WITH_TWEET_CONTENT" \\r
  --thread-id "THREAD_SHORT_ID" \\r
  --live-doc-id "LIVE_DOC_ID" \\r
  --accept-language LANG \\r
  --json\r
```\r
\r
**Query construction example:**\r
\r
> Please analyze the following tweets from @USERNAME and extract a writing style DNA document. Cover dimensions such as: tone, sentence structure, opening hooks, closing calls-to-action, frequently used words, hashtag strategy, emoji usage, and any other distinctive patterns.\r
>\r
> Account bio: [BIO]\r
>\r
> Tweets:\r
> [TWEET_1]\r
> [TWEET_2]\r
> ...\r
\r
Keep the query under 2000 characters. If tweet content is too long, include the most representative 10–15 tweets.\r
\r
#### Step 4: Save state\r
\r
Extract `thread_short_id` and `live_doc_short_id` from the JSON response fields `data.thread_short_id` and `data.live_doc_short_id`. Save for follow-up calls.\r
\r
---\r
\r
### Mode 2 — Content Creation\r
\r
**When:** User wants to create tweets, threads, or X long-form posts (with or without a style DNA).\r
\r
**Steps:**\r
\r
#### Step 1: Determine if style DNA is available\r
\r
- If Mode 1 was just run in this session → style DNA is already in the LiveDoc canvas, use follow-up mode\r
- If user provides a style DNA directly → include it in the query\r
- If user provides an account name → run Mode 1 first, then continue with Mode 2\r
- If no style DNA → proceed to style library selection (Step 1.5)\r
\r
#### Step 1.5: Brand Style Selection (new conversations only)\r
\r
**Only run this step when:**\r
- This is a **new conversation** (no `thread_short_id` in session), AND\r
- Mode 1 was NOT just run (i.e., there is no existing thread carrying style DNA context)\r
\r
**Skip this step entirely when:**\r
- `thread_short_id` already exists (follow-up) — `--ext` has no effect in follow-up mode\r
- Mode 1 was just run in this session — style context is already in the thread\r
\r
**1.5a. Check if user already specified a style:**\r
\r
If the user mentioned a style by name (e.g., "use my Bold Voice style") or pasted a style block, note it and skip to step 1.5d.\r
\r
**1.5b. Fetch the TWITTER style library (names only):**\r
\r
IMPORTANT: The style library output is very large (each Style DNA can be thousands of characters). Always use `--json` and extract only names/labels to avoid Bash tool output truncation. Never call `run_style_library.mjs` without `--json` for listing purposes.\r
\r
```bash\r
node felo-superAgent/scripts/run_style_library.mjs --category TWITTER --accept-language LANG --json | node -e "\r
const d=require('fs').readFileSync('/dev/stdin','utf8');\r
const j=JSON.parse(d);\r
const list=j.list||[];\r
const user=list.filter(s=>!s.recommended);\r
const rec=list.filter(s=>s.recommended);\r
if(user.length) { console.log('[Your styles]'); user.forEach((s,i)=>{ const labels=(s.content?.labels?.LANG||s.content?.labels?.en||[]).join(', '); console.log((i+1)+'. '+s.name+(labels?' — '+labels:'')); }); }\r
if(rec.length) { console.log('[Recommended styles]'); rec.forEach((s,i)=>{ const labels=(s.content?.labels?.LANG||s.content?.labels?.en||[]).join(', '); console.log((user.length+i+1)+'. '+s.name+(labels?' — '+labels:'')); }); }\r
if(!list.length) console.log('(No styles found)');\r
"\r
```\r
\r
Replace `LANG` with the user's language value (`zh`, `ja`, `ko`, `en`) in both `--accept-language` and inside the node script's `labels?.LANG` references.\r
\r
Always pass `--accept-language` matching the user's language (same value used for SuperAgent).\r
\r
**1.5c. Handle the result:**\r
\r
**If the list is empty:** Skip silently. Proceed to Step 2 without `--ext`.\r
\r
**If styles are available:** Output the COMPLETE list as plain text — every style returned by the API, grouped by type, numbered sequentially. NEVER use `AskUserQuestion` tool (it limits to 4 options and will silently drop styles). NEVER pre-select or filter styles on behalf of the user. Always append a "no preference" option as the last item. Then wait for the user's plain-text reply before proceeding.\r
\r
Example presentation (adapt language to match user's language):\r
```\r
Here are the available Twitter writing styles — choosing one will make the output more accurate:\r
\r
[Your styles]\r
1. My Bold Voice\r
\r
[Recommended styles]\r
2. elonmusk — Shitposting provocateur\r
3. naval — Pithy aphorism master\r
...(list ALL styles, do not omit any)\r
\r
0. No preference — use default style\r
```\r
\r
**1.5d. Build `--ext` from the chosen style:**\r
\r
After the user selects a style, fetch the full Style DNA for that specific style using `--json` and extract it in Node.js. Do NOT re-read the full text output — it will be truncated by the Bash tool.\r
\r
```bash\r
node felo-superAgent/scripts/run_style_library.mjs --category TWITTER --accept-language LANG --json | node -e "\r
const d=require('fs').readFileSync('/dev/stdin','utf8');\r
const j=JSON.parse(d);\r
const s=(j.list||[]).find(s=>s.name==='CHOSEN_STYLE_NAME');\r
if(!s){process.stderr.write('Style not found\
');process.exit(1);}\r
const labels=(s.content?.labels?.LANG||s.content?.labels?.en||[]).join(', ');\r
const dna=s.content?.styleDna||'';\r
const block='Style name: '+s.name+'\
Style labels: '+labels+'\
Style DNA: '+dna;\r
console.log(JSON.stringify({brand_style_requirement:block}));\r
"\r
```\r
\r
Replace `CHOSEN_STYLE_NAME` with the style name the user selected, and `LANG` with the language code.\r
\r
Pass the output JSON directly as the `--ext` value:\r
\r
```bash\r
--ext "$(node felo-superAgent/scripts/run_style_library.mjs --category TWITTER --accept-language LANG --json | node -e "\r
const d=require('fs').readFileSync('/dev/stdin','utf8');\r
const j=JSON.parse(d);\r
const s=(j.list||[]).find(s=>s.name==='CHOSEN_STYLE_NAME');\r
const labels=(s.content?.labels?.LANG||s.content?.labels?.en||[]).join(', ');\r
const dna=s.content?.styleDna||'';\r
const block='Style name: '+s.name+'\
Style labels: '+labels+'\
Style DNA: '+dna;\r
console.log(JSON.stringify({brand_style_requirement:block}));\r
")"\r
```\r
\r
If the user chose "no preference" (option 0), do NOT pass `--ext`.\r
\r
#### Step 2: Obtain live_doc_id\r
\r
Follow Constraint #4. If Mode 1 was already run, reuse the same `live_doc_id`.\r
\r
#### Step 3: Determine conversation mode\r
\r
| Condition | Mode | What to pass |\r
|-----------|------|--------------|\r
| No `thread_short_id` in this session (truly first call) | New conversation | `--live-doc-id` + `--skill-id twitter-writer` + `--ext` (if style chosen) |\r
| `thread_short_id` already exists in this session (all subsequent inputs, including after Mode 1) | Follow-up | `--thread-id` + `--live-doc-id` (NO `--ext`) |\r
| User says "new topic" / "start over" (clear `thread_short_id`) | New conversation | `--live-doc-id` + `--skill-id twitter-writer` + `--ext` (repeat Step 1.5) |\r
\r
#### Step 4: Call SuperAgent\r
\r
Replace `LANG` with the user's language: `en`, `zh`, `ja`, `ko`. See Constraint #6.\r
\r
**New conversation without style (no `thread_short_id`, user chose no preference or list was empty):**\r
\r
```bash\r
node felo-superAgent/scripts/run_superagent.mjs \\r
  --query "/twitter-writer ENRICHED_QUERY" \\r
  --live-doc-id "LIVE_DOC_ID" \\r
  --skill-id twitter-writer \\r
  --accept-language LANG \\r
  --json\r
```\r
\r
**New conversation with brand style:**\r
\r
```bash\r
node felo-superAgent/scripts/run_superagent.mjs \\r
  --query "/twitter-writer ENRICHED_QUERY" \\r
  --live-doc-id "LIVE_DOC_ID" \\r
  --skill-id twitter-writer \\r
  --ext '{"brand_style_requirement":"Style name: darioamodei\
Style labels: Thoughtful long-form essays\
Style DNA: # Dario Amodei (@DarioAmodei) Tweet Writing Style DNA\
\
## Style Overview\
Dario writes like a serious intellectual...（full content, do NOT truncate）"}' \\r
  --accept-language LANG \\r
  --json\r
```\r
\r
**Follow-up (`thread_short_id` already exists in session):**\r
\r
```bash\r
node felo-superAgent/scripts/run_superagent.mjs \\r
  --query "/twitter-writer USER_FOLLOW_UP" \\r
  --thread-id "THREAD_SHORT_ID" \\r
  --live-doc-id "LIVE_DOC_ID" \\r
  --accept-language LANG \\r
  --json\r
```\r
\r
**Query construction guidelines:**\r
\r
- Specify the content type: single tweet / thread / X long-form post\r
- Specify the topic clearly\r
- Include style DNA or reference account if available\r
- Default to 3 versions unless the user specifies otherwise\r
- Preserve the user's core intent; only add context and clarity\r
\r
**Query examples:**\r
\r
> Please write 3 versions of a tweet about [TOPIC] in the style of @USERNAME (style DNA above). Each version should feel distinct — vary the tone, structure, or angle.\r
\r
> Based on the style DNA extracted above, write a Twitter thread (5–7 tweets) about [TOPIC]. Start with a strong hook tweet.\r
\r
> Write an X long-form post about [TOPIC] following the writing style we analyzed. Aim for ~500 words.\r
\r
#### Step 5: Save state\r
\r
Extract `thread_short_id` and `live_doc_short_id` from the JSON response fields `data.thread_short_id` and `data.live_doc_short_id`.\r
\r
---\r
\r
## Mode Decision Logic\r
\r
```\r
User input\r
  │\r
  ├── Contains account name + "analyze / style / DNA / how does X write"\r
  │   OR: アカウント名 + "分析 / スタイル / DNA / どう書いている"\r
  │     → Mode 1 (Style DNA Extraction) — skip style library step\r
  │\r
  ├── Contains account name + "write / create / imitate / in the style of"\r
  │   OR: アカウント名 + "書いて / 作って / 風に / 真似て"\r
  │     → Mode 1 first → then Mode 2 (follow-up, skip style library step)\r
  │\r
  ├── Contains topic + "write / draft / tweet / thread / X post"\r
  │   OR: トピック + "書いて / ツイート / スレッド / Xの投稿"\r
  │     → Mode 2 directly\r
  │         └── New conversation?\r
  │               YES → Step 1.5: fetch TWITTER styles, present to user, wait for choice\r
  │               NO  → follow-up, skip style library step\r
  │\r
  └── Ambiguous (e.g., "help me with tweets" / "ツイートを手伝って")\r
        → Ask user: do they want to analyze an account's style, or create content?\r
```\r
\r
---\r
\r
## Complete Workflow Examples\r
\r
### Example A: Analyze an account's style\r
\r
```\r
User: "Analyze @paulg's tweet style"\r
```\r
\r
**Step 1:** Fetch tweets:\r
```bash\r
node felo-x-search/scripts/run_x_search.mjs --id "paulg" --user --tweets --limit 30\r
node felo-x-search/scripts/run_x_search.mjs --id "paulg" --user\r
```\r
\r
**Step 2:** Get `live_doc_id` (list or create)\r
\r
**Step 3:** Call SuperAgent (Mode 1 — no style library step):\r
```bash\r
node felo-superAgent/scripts/run_superagent.mjs \\r
  --query "/twitter-writer Please analyze the following tweets from @paulg and extract a writing style DNA document. Cover tone, sentence structure, opening hooks, closing CTAs, frequently used words, hashtag strategy, and emoji usage.\
\
Account bio: [BIO]\
\
Tweets:\
[TWEETS]" \\r
  --live-doc-id "LIVE_DOC_ID" \\r
  --skill-id twitter-writer \\r
  --accept-language en \\r
  --json\r
```\r
\r
**Step 4:** Save `thread_short_id` and `live_doc_short_id` from JSON response fields `data.thread_short_id` and `data.live_doc_short_id`.\r
\r
---\r
\r
### Example B: Create tweets with a reference style (Mode 1 → Mode 2)\r
\r
```\r
User: "Write 3 tweets about startups in @paulg's style"\r
```\r
\r
**Step 1:** Run Mode 1 to extract style DNA (same as Example A). Style library step is skipped because Mode 1 already establishes style context in the thread.\r
\r
**Step 2:** Reuse `live_doc_id` from Mode 1.\r
\r
**Step 3:** Follow-up call (continuing the same thread — `thread_short_id` from Mode 1, no `--ext`):\r
```bash\r
node felo-superAgent/scripts/run_superagent.mjs \\r
  --query "/twitter-writer Based on the @paulg style DNA extracted above, write 3 tweet variations about startups. Each should have a distinct tone and angle, within 280 characters." \\r
  --thread-id "THREAD_SHORT_ID" \\r
  --live-doc-id "LIVE_DOC_ID" \\r
  --accept-language en \\r
  --json\r
```\r
\r
---\r
\r
### Example C: Write a thread directly (Mode 2, new conversation, with style selection)\r
\r
```\r
User: "Write a Twitter thread about why most startups fail"\r
```\r
\r
**Step 1.5:** New conversation, no existing thread → fetch TWITTER styles (names only):\r
```bash\r
node felo-superAgent/scripts/run_style_library.mjs --category TWITTER --accept-language en --json | node -e "\r
const d=require('fs').readFileSync('/dev/stdin','utf8');\r
const j=JSON.parse(d);\r
const list=j.list||[];\r
const user=list.filter(s=>!s.recommended);\r
const rec=list.filter(s=>s.recommended);\r
if(user.length){console.log('[Your styles]');user.forEach((s,i)=>{const labels=(s.content?.labels?.en||[]).join(', ');console.log((i+1)+'. '+s.name+(labels?' — '+labels:''));});}\r
if(rec.length){console.log('[Recommended styles]');rec.forEach((s,i)=>{const labels=(s.content?.labels?.en||[]).join(', ');console.log((user.length+i+1)+'. '+s.name+(labels?' — '+labels:''));});}\r
if(!list.length)console.log('(No styles found)');\r
"\r
```\r
\r
Present to user:\r
```\r
Here are the available Twitter writing styles — choosing one will make the output more accurate:\r
\r
[Your styles]\r
1. My Bold Voice\r
\r
[Recommended styles]\r
2. darioamodei\r
\r
0. No preference — use default style\r
```\r
\r
User replies: `1`\r
\r
**Step 2:** Get `live_doc_id`.\r
\r
**Step 3:** New conversation with chosen style — extract full Style DNA via `--json` and pass as `--ext`:\r
```bash\r
node felo-superAgent/scripts/run_superagent.mjs \\r
  --query "/twitter-writer Write a Twitter thread (6–8 tweets) about why most startups fail. Start with a strong hook tweet that grabs attention. Each tweet should stand alone but flow naturally into the next." \\r
  --live-doc-id "LIVE_DOC_ID" \\r
  --skill-id twitter-writer \\r
  --ext "$(node felo-superAgent/scripts/run_style_library.mjs --category TWITTER --accept-language en --json | node -e "\r
const d=require('fs').readFileSync('/dev/stdin','utf8');\r
const j=JSON.parse(d);\r
const s=(j.list||[]).find(s=>s.name==='My Bold Voice');\r
const labels=(s.content?.labels?.en||[]).join(', ');\r
const block='Style name: '+s.name+'\
Style labels: '+labels+'\
Style DNA: '+s.content.styleDna;\r
console.log(JSON.stringify({brand_style_requirement:block}));\r
")" \\r
  --accept-language en \\r
  --json\r
```\r
\r
**Step 4:** Save `thread_short_id` and `live_doc_short_id` from JSON response fields `data.thread_short_id` and `data.live_doc_short_id`.\r
\r
---\r
\r
### Example D: Iterate on generated content (follow-up, no style step)\r
\r
```\r
User: "Make the 2nd tweet more humorous and add some emojis"\r
```\r
\r
Already have `thread_short_id` and `live_doc_id` from the previous call. This is a follow-up — do NOT fetch styles again, do NOT pass `--ext`.\r
\r
```bash\r
node felo-superAgent/scripts/run_superagent.mjs \\r
  --query "/twitter-writer Please revise the 2nd tweet generated above. Make the tone more humorous and lighthearted, and add appropriate emojis. Keep the original intent intact." \\r
  --thread-id "THREAD_SHORT_ID" \\r
  --live-doc-id "LIVE_DOC_ID" \\r
  --accept-language en \\r
  --json\r
```\r
\r
**Save** updated `thread_short_id` and `live_doc_short_id` from JSON response fields `data.thread_short_id` and `data.live_doc_short_id`.\r
\r
---\r
\r
### Example E: User specifies style by name\r
\r
```\r
User: "Write a tweet about AI trends using my 'Bold Voice' style"\r
```\r
\r
**Step 1.5a:** User already named the style → extract full Style DNA via `--json`. No need to ask the user again.\r
\r
**Step 3:** New conversation with that style:\r
```bash\r
node felo-superAgent/scripts/run_superagent.mjs \\r
  --query "/twitter-writer Write a tweet about AI trends" \\r
  --live-doc-id "LIVE_DOC_ID" \\r
  --skill-id twitter-writer \\r
  --ext "$(node felo-superAgent/scripts/run_style_library.mjs --category TWITTER --accept-language en --json | node -e "\r
const d=require('fs').readFileSync('/dev/stdin','utf8');\r
const j=JSON.parse(d);\r
const s=(j.list||[]).find(s=>s.name==='My Bold Voice');\r
const labels=(s.content?.labels?.en||[]).join(', ');\r
const block='Style name: '+s.name+'\
Style labels: '+labels+'\
Style DNA: '+s.content.styleDna;\r
console.log(JSON.stringify({brand_style_requirement:block}));\r
")" \\r
  --accept-language en \\r
  --json\r
```\r
\r
---\r
\r
### Example F: Style library is empty\r
\r
```\r
User: "Write a tweet about the new product launch"\r
```\r
\r
**Step 1.5b:** Fetch styles (names only):\r
```bash\r
node felo-superAgent/scripts/run_style_library.mjs --category TWITTER --accept-language en --json | node -e "\r
const d=require('fs').readFileSync('/dev/stdin','utf8');\r
const j=JSON.parse(d);\r
if(!(j.list||[]).length)console.log('(No styles found)');\r
else console.log((j.list||[]).map(s=>s.name).join('\
'));\r
"\r
```\r
\r
Output: `(No styles found)`\r
\r
**Skip silently.** Proceed directly without `--ext`:\r
```bash\r
node felo-superAgent/scripts/run_superagent.mjs \\r
  --query "/twitter-writer Write a tweet about the new product launch. Provide 3 versions with different tones." \\r
  --live-doc-id "LIVE_DOC_ID" \\r
  --skill-id twitter-writer \\r
  --accept-language en \\r
  --json\r
```\r
\r
---\r
\r
### Example G: User chooses no preference\r
\r
```\r
User: "Write a tweet about a new product launch"\r
```\r
\r
**Step 1.5:** Fetch styles, present list. User replies: `0` (no preference).\r
\r
Proceed without `--ext`:\r
```bash\r
node felo-superAgent/scripts/run_superagent.mjs \\r
  --query "/twitter-writer Write 3 tweets about a new product launch, each with a slightly different tone." \\r
  --live-doc-id "LIVE_DOC_ID" \\r
  --skill-id twitter-writer \\r
  --accept-language en \\r
  --json\r
```\r
\r
---\r
\r
## Error Handling\r
\r
| Scenario | Action |\r
|----------|--------|\r
| Account not found or no tweets returned | Inform user, suggest trying a different username or providing tweet samples manually |\r
| `FELO_API_KEY` not set | Stop and show setup instructions (same as `felo-superAgent` SKILL.md) |\r
| SuperAgent call fails | Check `live_doc_id` validity; retry once with the same parameters |\r
| Style library fetch fails | Log warning to stderr, skip silently, proceed without `--ext` |\r
| User asks for Mode 2 with no style DNA and no account | Proceed to Step 1.5 (style library selection) |\r
| User explicitly requests a new canvas | Create a new LiveDoc: `node felo-livedoc/scripts/run_livedoc.mjs create --name "Twitter Writer" --json` |\r
| Tweet content too long for query (>2000 chars) | Trim to the 10–15 most representative tweets; prioritize high-engagement ones |\r
\r
## References\r
\r
- [felo-superAgent SKILL.md](../felo-superAgent/SKILL.md) — SuperAgent calling conventions, `--ext` format, and style library script\r
- [felo-x-search SKILL.md](../felo-x-search/SKILL.md) — X/Twitter search commands\r
- [felo-livedoc SKILL.md](../felo-livedoc/SKILL.md) — LiveDoc management commands\r
- [Felo Open Platform](https://openapi.felo.ai/docs/)\r
- [Get API Key](https://felo.ai) (Settings → API Keys)\r