| name | x-writing-system-skill |
| description | Build a research-backed X writing system around a user draft. Use this skill when the user wants rewrite advice grounded in Matt Gray guidelines, their own top posts, topic winners, and trends. Then produce 5 LLM-authored post versions. |
X Writing System Skill
Use this skill when the user asks for:
- rewrite help for a draft post
- tactical writing feedback based on real X data
- a writing system that blends personal + market signals
Workflow
-
Ensure env is available (X_BEARER_TOKEN minimum).
-
Run the research brief command:
bun run x-search.ts advise \
--draft "<user draft>" \
--username <username> \
--performant-like-threshold 50 \
--topic-search-attempts 3
Use --draft-file if needed.
- Treat CLI output as evidence, then author the final answer with the model.
Required behavior
- Always incorporate Matt Gray writing guidelines as baseline.
- Always include personal top posts (last 30 days) when available.
- Always include topic research samples with metrics.
- Always include closest trends overlap output.
- Always give 3 specific recommendations.
- Always produce 5 distinct improved post versions authored by the model.
Do not output static template versions as the final result.
Final response format
Use this exact section order:
Closest Trending Topics
Topic Research (High-Performing Samples)
- Show concrete sample posts from topic research with metrics.
- Prefer >50 likes where available; if none qualify, show best available and say so explicitly.
3 Specific Recommendations
- Recommendation grounded in market research sample(s) + guideline(s)
- Recommendation grounded in personal top-performing post pattern(s)
- Recommendation grounded in both (with specific edits to apply)
5 Improved X Post Versions
Version 1
...
Version 2
...
Version 3
...
Version 4
...
Version 5
...
Guardrails
- Never print secret values.
- Never commit
.env.
- Treat CLI output as evidence, not final copy.
- Keep the 5 versions concise, original, and aligned with the user voice.
