Deterministic skills
Skills that give the same answer every time
These skills and modules compute their results with ordinary code, so your agent gets the same answer for the same input and can check it. Every one here carries its own tests and declares no network access.
70 components in 6 groups, each at the exact version the library release of October 9, 2026 serves.
Browse by job
- Statistics for comparing resultsCompare two models, two prompts or two harness setups with statistics that hold up: calibration, conformal intervals, confidence sequences, covariance shrinkage and kernel two-sample tests.
- Checks for agent workGive each unit of agent work one typed outcome, count the independent evidence behind a claim, and compare candidates only on the rows they all completed.
- Calculators for public statisticsRates, ratios and indices from published definitions, with their confidence bounds and the denominators written out.
- Records, schedules and budgetsCritical paths, earned value, shift rosters, appointment slots and retention dates as pure functions of JSON input.
- Algorithms that show their workAlgorithms that report each step: principal components without NumPy, sorting and union-find traces, rule based inference closures, collision clearance and resumable render plans.
- Geometry and mesh algorithmsPure Python geometry whose glTF output loads in Godot: triangulation, convex hulls, polyline simplification, polygon offsets, subdivision, curves and gears.
How your agent uses them
- Your agent searches Baltor for the step it is on and gets names, sizes and licences, never file bodies.
- It picks one component at an exact version and reads its file list, where every file names its SHA-256 checksum.
- It downloads exactly those files into your project, where your harness finds them, and checks each one against its checksum.
Why it matters
A model can explain a Brier score; a tested function computes it. Handing arithmetic, date math and parsing to code keeps model calls for the steps that need judgment, and the result can be checked against the component's own known answers.
Give your agent these components
Your harness downloads them through Baltor with Pro, $49 a month. Cancel any time from your account page.
More from Baltor
- Harness toolsTools your coding agent can call.
- Corner casesThe inputs that break code, ready to test.
- Stop starting from nothingFor a developer whose coding agent starts every task from nothing: give each step the guidance and reusable code it needs.
- More efficient operationGive each step only the context, files and tools it asks for, reuse code that already exists, and measure the effect on your own task.
- Harness FilesSkills, instruction files, tools, hooks and settings your harness picks up: what they are, how a harness gets them, and the setup for each harness.
- GalleryBrowse Baltor's original Godot shaders, procedural textures, meshes and Blender tools, each with the render its engine check recorded.
- Pi and Gemma 4 on Ollama CloudPi with Gemma 4 31B on Ollama Cloud searched the library, installed a skill, and a new Pi session loaded and used it.