Repository navigation
Releases
Kabo 1.0.26
Kabo 1.0.25
analyze-content-video:collect_media.py 的媒体交接改成 Web 宿主认得的写法(与 connector_driver.py 相同),Kabo Web 自己调用 collect_video_media、写入结果并续跑,模型不再经手暂存(#130)。本版基于 1.0.24 只加这一处修复。
Kabo 1.0.24
analyze-content-video:最终数字核对把带小数秒的时间点(0:07.866、0:07,866)整体当时间,不再把小数部分当成证据里找不到的数字而拒掉整份报告(#128)。本版基于 1.0.23 只加这一处修复。
Kabo 1.0.23
Video analysis: when its media result has not been brought into the run yet, the resume error now says to stage it first, so the diagnosis can reach its final checked report (#127).
Kabo 1.0.22
Kabo 1.0.22 is 1.0.21 plus the following.
What's changed
- Video analysis media collection:
analyze-content-videobuilds its single media call only through its own collection script. A failed call no longer falls back to per-operation media fetches, so the diagnosis reaches its final checked report (#125).
Kabo 1.0.21
Kabo 1.0.21 is 1.0.20 plus the following.
What's changed
- Remake edits after delivery:
remake-viral-videocan change a video Kabo already delivered (new words, caption style, a new background, an overlay or an inserted shot) withedit_remake_video, changing only the parts the user names (#80). - Step-by-step video projects: videos outside the remake formats are built with
start_video_project→make_video_step→finish_video_project, with the plan and cost shown first. New no-model steps:retime,crop,inpaint_textandlayout(#80). - Longer references: reference videos up to 10 minutes; transcripts up to 1,500 segments / 60,000 characters (#80).
- Contract: the directory MCP export is pinned to kabo main
8c8cfc0d(kabo#2022), 77 tools (#80).
Everything in 1.0.20 is included. OpenAI marketplace review remains separate.
Kabo 1.0.20
Kabo 1.0.20 is 1.0.19 plus the following.
What's changed
- Video analysis: the final figure check accepts counts rounded the way people write them (
1800万,1.8千万,2.6M+) and reads 千 / 百万 / 千万 as magnitudes (#120). Local measurement runs as its own command with a deadline (#115). - Reach-drop diagnosis: with only public snapshots, gives an approximate, age-aware drop verdict (#119).
- Web publishing: the pre test target no longer applies the production hold list (#121).
Everything in 1.0.19 is included. OpenAI marketplace review remains separate.
Kabo 1.0.19
Kabo 1.0.19 is 1.0.18 plus a fix to how the plugin's Skills are published to Kabo Web.
What's changed
- Web publishing: a Skill's declared
metadata.versionis used as its Web version when it is above the live one. The three Skills added in 1.0.18 (optimize-creator-brand,plan-content-calendar,price-creator-brand-deals) now declare 0.3.23, 0.1.25 and 0.4.13, so they can reach Web. Skill content is otherwise unchanged from 1.0.18.
Everything in 1.0.18 is included.
Assets
kabo-1.0.19.zip: the complete plugin package.kabo-1.0.19.zip.sha256: the package checksum.manifest.json: the source commit, file inventory and validation status.
Two independent builds produced identical ZIP files.
Kabo 1.0.18
Kabo 1.0.18 adds three Skills and brings over fixes made in the previous Skill source.
What's changed
- New Skills:
optimize-creator-brand(channel names, bios, usernames, profile visual spec, brand kit),plan-content-calendar(a reviewable four-week or date-range calendar) andprice-creator-brand-deals(line-by-line brand-deal pricing against benchmark bands). - Fixes brought over: TikTok photo and video posts get separate baselines and keep their original links (competitors, reach drop, publish timing); account reviews describe the inclusive maturity cutoff exactly; idea continuations can bind ordinary earlier conversation text; scripts route between creative drafts and native packages, keep an upstream selected idea, and hand finished-video requests to the host's media capability.
Everything in 1.0.17 is included.
Assets
kabo-1.0.18.zip: the complete plugin package.kabo-1.0.18.zip.sha256: the package checksum.manifest.json: the source commit, file inventory and validation status.
Two independent builds produced identical ZIP files.
Kabo 1.0.17
Kabo 1.0.17 lets video analysis get its media in one step that waits for it.
What's changed
analyze-content-video: media through one backend call. The transcript, frames and measurements (and the Instagram/TikTok metadata) now come from a singlecollect_video_mediacall that waits for every job, instead of the Skill submitting and polling each job itself. A diagnosis no longer stops while the media is still processing.
Everything in 1.0.16 is included.
Assets
kabo-1.0.17.zip: the complete plugin package.kabo-1.0.17.zip.sha256: the package checksum.manifest.json: the source commit, file inventory and validation status.
Two independent builds produced identical ZIP files.