15 projects
vetbox
Agentic validation sandbox: policy-file validation of agent outputs in safety-critical domains, with hash-chained traces and deterministic replay
inflame-trace
Where does your LLM inference latency live? vLLM/TGI/Triton traces in, flamegraph-style HTML out.
ifc-spf
Read IFC-SPF (STEP) files with the standard library — instances, property sets, quantities and units, no geometry kernel.
portover
Migrate one config dialect to another, one directive at a time: pip->uv, Jenkins/Travis/CircleCI/GitLab CI/Azure Pipelines/Bitbucket/Buildkite/Drone/Woodpecker->GitHub Actions, flake8->ruff
qto-takeoff
Open quantity takeoff from IFC, with pluggable cost classification (UniFormat, MasterFormat, NRM, OmniClass, định mức)
snapback-cli
Undo for your machine: snapshot working tree, databases, and container volumes before a risky command; roll back with one command.
agentbox-runtime
Policy-sandboxed runtime for AI agent processes: enforce a policy file, capture a tamper-evident trace, deterministically replay any run.
whybig
Why is this so big? Treemap + ranked culprits and fixes for Docker images, Python wheels, model checkpoints, JS bundles, and git repos.
git-split-cli
Carve a messy multi-file Git branch into clean, reviewable commits
actdbg
Run GitHub Actions workflows locally with a real debugger: breakpoints, step-through, shell at the failure point, re-run one step without re-running the job.
whalint
A tiny, zero-dependency Dockerfile linter where every rule is one small file
callscape
Point it at a codebase, get a browsable static HTML call-graph map. Built on tree-sitter.
pytest-flakeradar
Flaky-test radar for pytest: per-test flake rate across runs, with a FLAKY.md leaderboard your CI commits back
judge-audit
Audit the reliability of an LLM-as-judge eval pipeline: agreement, bias, drift, calibration.
lyapmon
Online Lyapunov-drift monitor for ML retraining loops: alert when the loop trends unstable, before eval metrics show it.
