360 Secure SkillHub
360 安全技能中心

instagram-carousel

该 Skill 将用户主题或现成内容映射为 Canvora 轮播参数,强调钩子优先、每页一个要点、明确的品牌/风格/CTA,并指导异步轮询、按 sortOrder 排序、处理 partial 结果及用 edit/variations 修订。设计执行由 Canvora 完成,不提供 Instagram 发布或逐页美术指导。

分类:营销
canvora
热度 2.73
最近更新 2026-10-28 19:27:35

安装

发给 Agent
帮我安装这个技能:npx secure-skills-cli add canvora/canvora-instagram-carousel
兼容性
通用兼容
联网要求
需要联网
数据出网
必须出网
Metadata
name
instagram-carousel
version
1.1.0
description
Create high-performing Instagram carousels with Canvora — right format, right slide count, hook-first content shaping, and the correct refine workflow. Use when the user wants an Instagram carousel, multi-slide IG post, swipe post, or educational/listicle content for Instagram.
homepage
https://canvora.ai/for-agents
metadata
{"openclaw":{"emoji":"🎠","requires":{"bins":["canvora"],"env":["CANVORA_API_KEY"]}}}

Instagram Carousel Creation

Map the user's intent onto the right Canvora levers. Canvora's carousel pipeline owns the design execution — layout, composition, imagery, visual variety per slide. Your job is the content and the parameters. Setup, credits, and general CLI usage: see the canvora skill.

Intake — ask only what you can't infer

Usually you already know the topic. Then you need at most two more things:

  1. How many items? ("5 tips" → 5, plus a cover if they want one). If they don't say, use 6.
  2. Whose brand? If a client/company is named → canvora brands --json → --brand. If they have no kit, don't ask again.

Anything else — style, language, CTA — infer it and state your assumption in one line when you report back. Do not run a questionnaire.

The rule that matters

Don't art-direct slide by slide. Do not prescribe per-slide layouts, color values, or illustration details; over-specified briefs fight the tuned format pipeline and produce worse, samey output.

But do carry the user's stated look. If they said "black and white with big type" or "hand-drawn", pass it through as one line of overall direction. It is honoured, and dropping it is the more common and more damaging mistake. The line is: overall aesthetic = yes, slide-by-slide instructions = no.

Parameter mapping (your actual job)

User intent Lever
Any IG carousel --format instagram_carousel_portrait (1080×1350 — outperforms square for reach). Square (instagram_carousel) only if they ask for it.
"N tips/mistakes/steps" --slides instagram_carousel_portrait=N (or N+1 with a cover). Never leave a promised count unmatched. No count promised → 6.
A named client or brand canvora brands --json → --brand <uuid>. Mandatory when a kit exists. Add --logo to actually stamp the logo.
Tone words --style: educational/punchy → bold; clean/professional → modern or minimal; luxury/coaching → elegant; consumer/fun → playful.
No tone signal at all --style modern. Note: omitting the flag does NOT auto-pick — it pins modern anyway. Choosing deliberately is free.
"End with a follow/save CTA" --cta (final slide becomes the call to action)
Exact wording required (quotes, taglines) --input "..." --exact
User gave a topic only --idea (Canvora develops the substance)
User gave finished content --input (wording respected)

Content shaping (what goes IN the brief — copy craft, not design)

  • Hook first: phrase the idea as the specific promise the cover should carry — "5 client-content mistakes that kill engagement", not "content mistakes". Numbers and specificity win.
  • One idea per item: if using --input, structure it as a clean list (item = one point, ≤25 words). The pipeline maps items to slides.
  • End intent: if they want a save/follow CTA, use --cta rather than describing it in prose.
ID=$(canvora generate \
  --idea "5 client-content mistakes that kill engagement — for freelance social media managers" \
  --format instagram_carousel_portrait --slides instagram_carousel_portrait=6 \
  --brand $BRAND_ID --logo --style bold --cta | head -1)
canvora status "$ID" --json    # repeat every ~30s until completed
  • A carousel is always 3+ outputs, so fire-and-poll — never --wait, which your harness usually kills at 120s while the generation carries on server-side. Never re-run generate after a timeout; poll status. Detail in the canvora skill's "Long runs" section.
  • State the cost in the same message as the run — 15 credits per slide, so the 6-slide carousel above is 90. Announce it, don't ask permission. Confirm the number only when you picked it (the 6 default), not when they said "5 tips".

After generation

  • Sort outputs by sortOrder before presenting or posting — outputs can complete out of order.
  • partial status? Some slides were blocked or failed and those credits were refunded. Say so, and offer to re-run the missing slides — don't hand over a 4-slide carousel that was meant to be 6 without explaining.
  • Different look wanted? canvora variations <id> — fresh takes from the pipeline (15 credits per slide). Do NOT regenerate with a more prescriptive brief.
  • One slide needs a fix? canvora edit <outputId> --prompt "..." --yes — targeted, design preserved, 3 credits. Always the first thing to try.

Quality checklist

  • Portrait format; slide count matches the promised item count
  • Brand kit applied when one exists (--brand, plus --logo if they want the mark)
  • Brief carries hook + content items + the user's stated look — no invented art direction
  • Style chosen deliberately, not left to default
  • Outputs sorted by sortOrder; partials reported
  • Revisions via edit (3 cr) or variations, never regeneration