AI Decision System
Replayable, offline-verifiable records for AI-assisted go/no-go decisions: experiment promotions, typed model answers and human-approved actions.
翻译准备中
个人 GitHub 账号下公开仓库的所有者可以登录并验证所有权,自行管理介绍,无需管理员批准。
登录并生成验证文件,在 30 分钟内将其提交到此仓库的默认分支。其他分支或分叉仓库中的文件无效。然后在下方检查文件,即可无需管理员批准直接管理项目。请保留文件以维持管理权限。
如需登录,将前往登录页面。登录后请回到此处,再次点击所选的验证按钮。
在项目文件夹中运行 /nomorevibe verify,即可验证所有权并自行管理介绍。
一分钟完成验证Code package dependency analyzer
翻译准备中
取自仓库 README
正在加载完整 README。 以下为已保存的摘录。
Code deps health tracker.
Parses SemanticDB or jdeps output
into the codeps export format,
then emits a flat metrics report:
SCC facts with optional budgeted cut analysis, per-node exposed-surface metrics
(ports/mutPorts/exposure/dependentsPerPublicPort) and declaration-surface
encapsulation —
over the package graph, or the file graph of the packages you select.查看完整 README ↗Code deps health tracker.
Parses SemanticDB or jdeps output
into the codeps export format,
then emits a flat metrics report:
SCC facts with optional budgeted cut analysis, per-node exposed-surface metrics
(ports/mutPorts/exposure/dependentsPerPublicPort) and declaration-surface
encapsulation —
over the package graph, or the file graph of the packages you select.
Your local build already produces everything codeps needs: a compiler with SemanticDB enabled emits
.semanticdb files alongside your compiled classes, and the JDK ships jdeps — codeps just reads
that existing output. No extra tooling to install.
Requires a JDK (11+).
# macOS or Linux (Homebrew)
brew install sake92/tap/codeps
# Scala: compile with SemanticDB enabled
scala-cli compile --server=false --semanticdb -d classes src/
# Configure codeps to read the compiler output, then analyze it
mkdir -p .codeps
cat > .codeps/config.yaml <<'EOF'
projects:
app:
root: .
source: semanticdb
inputs: [classes/META-INF/semanticdb]
skip-tests: true
EOF
codeps status
One status run writes package and file reports when the selected source
supplies file data. See the jdeps setup guide
for Java input.
For any other ecosystem, produce the codeps export format
with a tool of your choice (madge, pydeps, go list, ...), configure it as an
export source, then run status — codeps never parses that source code itself:
madge --json src | jq '... shape it into the codeps export format ...' > deps.json
cat > .codeps/config.yaml <<'EOF'
projects:
app:
root: .
source: export
inputs: [deps.json]
EOF
codeps status
The metrics are language-agnostic once the node/edge list carries per-node isExposed/ports/mutPorts
(see the metrics report) —
a TS/Python extractor only has to emit the same generic JSON shape.
Start with bounded findings from the latest cached report, then inspect selected subjects:
codeps findings --project app --limit 10 --format json
codeps inspect-cycle --project app --id scc:com.example.orders
codeps inspect-node --project app --id com.example.orders
Use ids returned by findings and preserve the selected project and scope. Check both
omitted and sourceFindingsOmitted before treating the selection as exhaustive.
The CLI reference covers finding
filters, file scope, and inspection. Cycle-cut analysis is available in the core library;
CLI reports have cutAnalysis.status: notRequested.
Developing codeps itself requires deder (brew install sake92/tap/deder):
deder exec -t test # run all tests
deder exec -t run -m cli status
The website is built with flatmark:
./scripts/build-docs.sh # outputs docs/_site/
Deployed to GitHub Pages on every push to main (see .github/workflows/ghpages.yml).