CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

draft-review-posts

Draft social posts and vendor feedback from a local model eval's conclusion pack, and publish only after the user approves the exact content.

Quality

82%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

Draft Review Posts

Use this after write-conclusion-pack has written report.md in a run directory.

What makes these posts work

The posts people share about local models answer "does this run on hardware like mine, and is it good enough to use". They lead with the machine, the model, the engine and quantization, and a concrete number; they show the recipe; and the ones that last are honest about limits. A surprising result on ordinary hardware travels further than a big number on expensive hardware.

Write in the user's own voice; read a few of their past posts first if an X account is connected. Plain sentences, no hype words, no emoji walls.

Drafts to produce

Save each to <run_dir>/posts/ and show them in the chat:

  1. X post: one post under 280 characters plus the speed chart. Machine, model, quantization, engine, the headline number, one surprising finding.
  2. X thread (4 to 6 posts): the recipe (exact command or settings), the capability results with sample counts, the caveats worth knowing, the limitations, and a line for a link to the full report.
  3. Chinese short post (for Xiaohongshu, Jike or similar): the same facts in natural Chinese, with the chart.
  4. Vendor feedback (vendor-feedback.md, never published): what is worth promoting, what to fix with the evidence, and the follow-up experiments that confirmed it.

Every number in a draft must appear in report.md. Do not add links you do not have; leave a placeholder.

Disclosure

If the review is paid or done in partnership with the model's vendor, the post must say so (#ad, "Sponsored", or "合作"). If you do not know, ask the user before drafting.

Publishing

Posting goes through the user's connected accounts (X, LinkedIn, Reddit). holaOS blocks posting tools until the user approves the exact text, so:

  1. Show the final text and attached image, and ask for approval with ask_user_question.
  2. Publish exactly the approved content. If the user edits it, show the new version and ask again.

For platforms without a connected integration (for example Xiaohongshu), give the user the text and the image files to post themselves.

Repository

holaboss-ai/holaOS

Path

runtime/api-server/src/embedded-capabilities/local-model-review/skills/draft-review-posts/SKILL.md

Last updated
First committed

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.