| name | create-x-content |
| description | Draft 2–5 voice-tuned X post variants from a free-form brief, each with a distinct framing (how-to, mechanism, problem-first, hype, contrarian…), and self-check each against a personal voice guide's banned phrases and format rules before returning — so posts sound like the user, not like an AI. The agent does the drafting; a bundled deterministic linter enforces banned-phrase / length / hashtag rules. Invoked standalone or by social-kit. |
| metadata | {"version":"1.0.1","category":"content","type":"capability"} |
Create X Content
The drafting is yours (the agent is the model). This skill is mostly instructions plus
a tiny deterministic linter (lint_post.py) that mechanically checks each variant.
When to use
- "Write X posts about [topic/launch/build]."
- When you want several genuinely different angles on one idea.
- As the X arm of
social-kit. Pairs with generate-voice-guide and graphics-studio.
How to draft
- Resolve the voice guide: explicit path → stored config / Memory → default location →
else prompt to generate one (
generate-voice-guide) or proceed with a warned generic
default. Never silently skip it — generic AI tweets are the failure mode.
- Parse the brief; decide variant count by richness (2 = single-angle; 5 = mechanism-rich).
- Draft each variant under a distinct framing the voice guide's hook patterns support
(simple-howto, howto-plus-mechanism, problem-first, hype, mechanism-breakdown,
ecosystem-map, contrarian, personal-experience). Quote the brief's specifics verbatim
(numbers, tool names, prices) — that's the "meat". For rich briefs with a field-notes
hook pattern, include one long-form variant.
- Self-check each variant with the linter, then your own voice-match judgement. Rewrite
≤2× or drop. Prefer fewer strong variants over many weak ones.
- Save one file per variant (
variant-<letter>-<framing>.md) with frontmatter
(platform: x, format: short|long, framing, status: draft) and just the post text.
Linter
python3 ${SKILL_DIR}/scripts/lint_post.py --file ${WORKSPACE}/variant-a-howto.md \
--voice-guide ${WORKSPACE}/voice-x.md --format auto
| Flag | Meaning |
|---|
--file | Post file (default stdin; frontmatter is stripped). |
--banned | Comma-separated extra banned phrases (added to the built-in list). |
--voice-guide | Voice-guide file; pulls its "Banned phrases" section. |
--format | short (<280) / long (280–4000) / auto. |
--max-hashtags | Hashtag ceiling (default 2). |
Emits JSON: {format, char_count, length_ok, banned_hits, hashtag_count, hashtags_ok, link_count, has_concrete_token, pass}. pass=false (banned hit, length, or hashtags) →
rewrite or drop. has_concrete_token=false is your cue to check the variant has real "meat".
python3 ${SKILL_DIR}/scripts/lint_post.py --help lists all flags.
Outputs
One markdown file per variant + a short summary of framings used and a suggested next step
(e.g. pair with a graphics-studio visual). No index file.
Credentials / env
- Required: none. The agent is the drafting engine; no script needs a key.
- Optional: none. No paid service is applicable — this is pure LLM drafting with no
external-data step. A missing voice guide degrades gracefully (prompt or warned-generic).
Notes & edge cases
- Default to fewer variants — two strong beats five weak. Framings must genuinely differ.
- Never silently skip the voice guide.
- Mirrors
create-linkedin-content; the difference is the platform defaults (length, no
arrow-bullet / "why this matters" requirement here).
