Skip to content
Back to skills

Daniel X Writer

ASecurity

Use for concise Chinese X/Twitter posts in Daniel's first-person voice, mixing human tension, numerical evidence, explicit decision rules, and a sharp reflective ending.

  • 92 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added September 22, 2026
ai-agentspythonbash

Security analysis

A100/100

Pro scans all 5 files and shows the line behind each finding

Scanned September 22, 2026

npx -y skills add aAAaqwq/AGI-Super-Team --skill daniel-x-writer --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Daniel X Writer?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Daniel X Writer
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/aaaaqwq-daniel-x-writer-agi-super-team/badge)](https://www.skillsdirectory.com/skills/aaaaqwq-daniel-x-writer-agi-super-team)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: daniel-x-writer
description: "Use for concise Chinese X/Twitter posts in Daniel's first-person voice, mixing human tension, numerical evidence, explicit decision rules, and a sharp reflective ending."
metadata: {"runtime_effects":"local-read-only; does not publish or call network services"}
---

# Daniel X Writer

Turn raw analysis into one copy-ready X post that sounds lived-in, not generated.

## Draft workflow

1. Identify the one thing the reader should remember. Drop secondary analysis.
2. Open with a real human impulse, doubt, mistake, or tension. Prefer “我当时想……” over a generic market summary.
3. Personify the Agent once as a calm counterforce: it may press a hand down, ask one hard question, or refuse to validate a hunch. Do not turn it into a cute mascot or an oracle.
4. Keep at most three hard observations. Preserve the strongest prices, volumes, ratios, or measured facts exactly.
5. Convert the decision into a rule: state the trigger and, when relevant, the condition that proves the idea wrong.
6. End with a short human insight. Make the point about discipline, thinking, or responsibility—not about being right.
7. Compress and verify before responding.

## Voice rules

- Sound direct, self-aware, and slightly sharp.
- Mix emotion with evidence; never replace evidence with drama.
- Use short sentences and 3–5 visual paragraphs.
- Use no more than one metaphor and one quoted Agent line.
- Prefer concrete verbs: `按住`, `逼我写清`, `砸出`, `认错`.
- Let numbers carry credibility. Do not explain every indicator.
- Keep the Agent subordinate to human responsibility: it challenges the decision; it does not own the trade.

Read [references/style-examples.md](references/style-examples.md) when matching tone, selecting a structure, or diagnosing why a draft feels generic.

## X length contract

- Default to one ordinary X post unless the user explicitly asks for a thread or long post.
- Treat 280 weighted characters as the hard ceiling. Target 240–260 to leave editing room.
- Never solve an over-limit draft by silently turning it into a thread.
- Remove headings, repeated caveats, full evidence chains, and decorative adjectives before removing decisive facts.
- When a draft is stored in a local UTF-8 text file, run:

```bash
python3 scripts/check_x_length.py /absolute/path/to/draft.txt
```

The checker deliberately overcounts non-ASCII characters and URLs conservatively. If it reports `over`, shorten the post and rerun it. Treat the final X composer as authoritative when available.

Run its local regression tests after changing the checker:

```bash
python3 -m unittest scripts/test_check_x_length.py
```

## Truth and risk rules

- Separate what happened before the decision from what happened afterward.
- Never turn a subjective scenario weight into a model probability or win rate.
- Never invent entry price, position size, leverage, profit, backtest results, or data provenance.
- For financial content, use a brief historical-recap boundary when readers could mistake the post for a live signal. Do not bury the post under legal boilerplate.
- Prefer “提高可信度” to “证明”,and “偏空依据” to “必跌信号”.
- If the source is internally inconsistent, preserve the core story but flag the conflict instead of repairing facts silently.

## Output contract

- Return exactly one copyable text block by default.
- Do not add analysis, character-count commentary, alternative versions, hashtags, or a publishing plan unless requested.
- Keep the user's language and natural code-switching. Use `Agent`, `BB`, prices, and ticker notation only where they feel native.
- Before returning, check: human tension, one Agent moment, no more than three facts, trigger, invalidation when available, sharp ending, and length under the requested limit.

Files in this skill

  • SKILL.md3.8 KB
  • agents/openai.yaml262 B
  • references/style-examples.md2.3 KB
  • scripts/check_x_length.py2.2 KB
  • scripts/test_check_x_length.py1.8 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.