Install any skill in seconds. Free to start, no credit card required.
Get Started Free →用于发布 ccLoad 新版本,自动提交未提交改动并推送本地领先的 master,按固定版本通道计算并发布 Tag,等待 GitHub Actions,以及验证 GitHub Release 和对应通道的容器镜像。Beta 固定沿用最近稳定版的主版本和次版本;只有显式 stable 发布才允许修改次版本。
.claude/skills/caidaoli-ccload-release/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | — | 2 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 138% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -13% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 37% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 37% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 47% | 0% |
通过唯一的 Tag 驱动 .github/workflows/release.yml。不要手动创建 Release、手动触发发布工作流或单独发布容器镜像。
只要求计算版本或预检时,执行下方对应的 dry-run 并报告结果,不进入 publish。用户要求发布或单独调用本技能(可带通道参数)才适用自动提交、推送和发布授权;编辑技能文件不执行发布。
$ccload-release、/ccload-release、ccload-release beta、ccload-release preview:发布下一个 Beta。ccload-release stable:发布稳定版。必须有显式 stable 参数;禁止根据语气猜测。beta、preview、stable。Tag 形状固定:
vX.Y.Z-beta.NvX.Y.ZvX.Y.S 的主版本 X 和次版本 Y,禁止因 feat 或 breaking change 生成 vX.(Y+1).0-beta.N 或 v(X+1).0.0-beta.N。vX.Y.(S+1)-beta.1。vX.Y.P-beta.N 时,只检查该 Beta Tag 之后的新提交:feat、! 或 BREAKING CHANGE)保持 X.Y.P,发布 vX.Y.P-beta.(N+1)。feat、! 或 BREAKING CHANGE)只把 patch 加一并重置序号,发布 vX.Y.(P+1)-beta.1。stable 发布才从最近稳定版后的全部提交计算标准 SemVer:breaking change 增加主版本,feat 增加次版本,其余增加 patch。Beta 永远不修改主版本或次版本。git status 和完整 diff。工作区有改动时,根据实际改动推导一个单行 Conventional Commit subject;必须准确反映最高语义版本影响,禁止使用掩盖功能变更的通用 chore。不要为提交或推送重复询问用户。beta。工作区干净时运行:bash bash .agents/skills/ccload-release/scripts/release.sh beta --dry-run
工作区有改动时传入推导出的提交 subject:
bash bash .agents/skills/ccload-release/scripts/release.sh beta --dry-run --commit-message 'feat(scope): summary'
稳定版把 beta 改为 stable。
master 和发布;目标符合参数契约时直接继续。bash bash .agents/skills/ccload-release/scripts/release.sh beta --publish
工作区有改动时必须复用 dry-run 的同一个 subject:
bash bash .agents/skills/ccload-release/scripts/release.sh beta --publish --commit-message 'feat(scope): summary'
稳定版把 beta 改为 stable。脚本会自动 git add -A、创建提交、非强制推送 master,等待该提交对应的 Test workflow 全部成功,确认远端仍精确一致后再创建并推送 annotated Tag。本地已有未推送提交时不创建额外提交,直接推送并等待 CI。
master 修订、目标 Tag、GitHub Release URL 和 Actions 结果。稳定版报告 ghcr.io/caidaoli/ccload:<tag> 和 ghcr.io/caidaoli/ccload:latest(该稳定版为 SemVer 最高版本时 beta 别名同步指向它,一并报告);Beta 报告 ghcr.io/caidaoli/ccload:<tag> 和 ghcr.io/caidaoli/ccload:beta。master。本地 master 可与 origin/master 相等或领先;远端领先或双方分叉时停止,禁止自动 pull、merge、rebase 或 force-push。Test workflow 是提交级唯一发布门禁,必须覆盖后端测试、Web 验证、构建、lint 和 PostgreSQL 集成测试。发布脚本必须按完整 commit SHA 找到并等待该 workflow;找不到、被取消或失败时不得创建 Tag。Test workflow 成功后都必须重新 fetch 并确认本地 HEAD 等于 origin/master,才能创建 Tag。CI 失败时保留已经推送的提交,不自动回滚。latest;Beta 必须同时打精确版本 Tag 和 beta;精确标签与对应浮动别名必须指向同一镜像摘要。beta 别名指向全渠道 SemVer 最高版本:稳定版发布时若不存在更高的 Beta Tag,beta 同步推进到该稳定版;存在更高 Beta 时保持不动,禁止降级。.github/workflows/release.yml 的 Tag 校验,也不创建 beta、latest 这类浮动 Git Tag;它们只允许作为 GHCR 镜像别名由发布工作流管理。修改发布脚本后运行:
bashbash .agents/skills/ccload-release/scripts/release.sh --self-test
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 13,334 | 13,815 | +4% | 1 | 1 | 0% | 296 | 2,115 | +615% | 0 | 0 | — |
case-02 | fail→fail | 10,937 | 19,425 | +78% | 1 | 1 | 0% | 201 | 2,084 | +937% | 0 | 0 | — |
case-03 | fail→fail | 21,309 | 30,455 | +43% | 1 | 1 | 0% | 355 | 1,936 | +445% | 0 | 0 | — |
case-04 | fail→pass | 14,812 | 11,619 | -22% | 1 | 1 | 0% | 1,410 | 3,359 | +138% | 0 | 0 | — |
case-05 | pass→pass | 15,799 | 10,235 | -35% | 1 | 1 | 0% | 1,760 | 2,836 | +61% | 0 | 0 | — |
case-06 | pass→pass | 12,591 | 12,869 | +2% | 1 | 1 | 0% | 1,405 | 2,740 | +95% | 0 | 0 | — |
case-07 | fail→pass | 20,839 | 8,948 | -57% | 1 | 1 | 0% | 2,428 | 2,114 | -13% | 0 | 0 | — |
case-08 | fail→pass | 12,381 | 5,434 | -56% | 1 | 1 | 0% | 1,717 | 2,357 | +37% | 0 | 0 | — |
case-09 | pass→pass | 19,215 | 5,775 | -70% | 1 | 1 | 0% | 1,998 | 2,358 | +18% | 0 | 0 | — |
case-10 | pass→pass | 15,864 | 11,290 | -29% | 1 | 1 | 0% | 1,520 | 2,416 | +59% | 0 | 0 | — |
case-11 | fail→pass | 15,066 | 2,534 | -83% | 1 | 1 | 0% | 1,400 | 1,916 | +37% | 0 | 0 | — |
case-12 | fail→fail | 16,431 | 19,657 | +20% | 1 | 1 | 0% | 1,595 | 2,428 | +52% | 0 | 0 | — |
case-13 | pass→pass | 19,768 | 11,196 | -43% | 1 | 1 | 0% | 1,939 | 2,518 | +30% | 0 | 0 | — |
case-14 | fail→pass | 17,721 | 12,744 | -28% | 1 | 1 | 0% | 1,876 | 2,754 | +47% | 0 | 0 | — |
case-15 | fail→fail | 15,844 | 17,768 | +12% | 1 | 1 | 0% | 1,788 | 2,333 | +30% | 0 | 0 | — |
case-16 | pass→pass | 17,152 | 11,399 | -34% | 1 | 1 | 0% | 1,894 | 2,680 | +41% | 0 | 0 | — |
case-17 | pass→pass | 14,543 | 11,137 | -23% | 1 | 1 | 0% | 1,473 | 2,530 | +72% | 0 | 0 | — |
case-18 | pass→pass | 41,468 | 9,251 | -78% | 1 | 1 | 0% | 1,736 | 2,360 | +36% | 0 | 0 | — |
case-19 | pass→pass | 12,689 | 11,303 | -11% | 1 | 1 | 0% | 1,745 | 2,462 | +41% | 0 | 0 | — |
case-20 | pass→fail | 12,683 | 13,606 | +7% | 1 | 1 | 0% | 1,543 | 2,180 | +41% | 0 | 0 | — |
case-21 | pass→fail | 28,550 | 25,396 | -11% | 1 | 1 | 0% | 1,451 | 2,080 | +43% | 0 | 0 | — |
case-22 | pass→pass | 13,712 | 11,339 | -17% | 1 | 1 | 0% | 1,175 | 2,523 | +115% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 15 counted toward the lift figure. The other 7 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +14 percentage points is the difference between those two pass rates over the 15 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 9/2/2026 | +35% |
Other measured skills in the registry, with their headline benchmark lift.