pymc-modeling
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/pymc-labs--pymc-modeling)HTML
<a href="/listings/pymc-labs--pymc-modeling"><img src="/api/listings/pymc-labs--pymc-modeling/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
Local checkout
From the repository root, use the [Skills CLI](https://github.com/vercel-labs/skills):
npx skills add .
Choose the skills and agents you want.
For a local Claude Code managed bundle, run these commands from the repository root inside Claude Code:
/plugin marketplace add ./ /plugin install pymc-modeling@pymc-modeling
This local installation does not require the renamed marketplace to be published. Use a clean checkout for this installation: Claude Code can copy untracked files too, including a maintainer's `.pixi/` environments.
Or copy individual directories from `skills/` into your agent's supported skills location. Keep the whole directory, including references, any scripts and `LICENSE`. No custom installer or running service is required.
Published repository (publication required)
Use the following commands only after the matching skills and marketplace manifest have been published to `pymc-labs/pymc-modeling`. The published manifest must name both the marketplace and its plugin `pymc-modeling`. Until then, use the local-checkout instructions above. Local validation does not establish remote availability or successful remote plugin installation.
For the Skills CLI:
npx skills add pymc-labs/pymc-modeling
For a Claude Code managed bundle:
/plugin marketplace add pymc-labs/pymc-modeling /plugin install pymc-modeling@pymc-modeling
Choose either the plugin or the Skills CLI/manual installation, not both, to avoid duplicate skills.
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
.claude-plugin/marketplace.json · Manifest · 22 lines
View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)
- {
- "name": "pymc-modeling",
- "version": "6.4.0",
- "owner": { "name": "PyMC Labs" },
- "metadata": { "description": "Agent skills for PyMC, PyTensor, prior elicitation, ArviZ and MLflow tracking." },
- "plugins": [
- {
- "name": "pymc-modeling",
- "version": "6.4.0",
- "description": "Bayesian modeling, prior elicitation, ArviZ diagnostics, PyTensor workflows and MLflow tracking.",
- "source": "./",
- "strict": false,
- "skills": [
- "./skills/pymc-modeling",
- "./skills/prior-elicitation",
- "./skills/arviz-diagnostics",
- "./skills/pytensor-workflows",
- "./skills/pymc-mlflow"
- ]
- }
- ]
- }
README.md · README · 175 lines
View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)
- # pymc-modeling
- Agent skills for Bayesian modeling with PyMC, PyTensor and ArviZ, and experiment
- tracking with MLflow. Each skill is a folder of instructions and focused references in the
- [Agent Skills format](https://agentskills.io/specification).
- ## Skills
- | Skill | Use it for |
- |---|---|
- | [pymc-modeling](skills/pymc-modeling/SKILL.md) | Model specification, inference, predictions, specialized model families and pymc-extras workflows |
- | [prior-elicitation](skills/prior-elicitation/SKILL.md) | Prior selection, elicitation, predictive checks and shrinkage |
- | [arviz-diagnostics](skills/arviz-diagnostics/SKILL.md) | MCMC diagnostics, predictive checks, LOO and model comparison |
- | [pytensor-workflows](skills/pytensor-workflows/SKILL.md) | Symbolic graphs, shapes, compilation, gradients and custom Ops |
- | [pymc-mlflow](skills/pymc-mlflow/SKILL.md) | MLflow experiment tracking, DataTree artifacts, PyMC-Marketing autologging and MMM persistence |
- ## Install
- ### Local checkout
- From the repository root, use the [Skills CLI](https://github.com/vercel-labs/skills):
- ```bash
- npx skills add .
- ```
- Choose the skills and agents you want.
- For a local Claude Code managed bundle, run these commands from the repository
- root inside Claude Code:
- ```text
- /plugin marketplace add ./
- /plugin install pymc-modeling@pymc-modeling
- ```
- This local installation does not require the renamed marketplace to be published.
- Use a clean checkout for this installation: Claude Code can copy untracked
- files too, including a maintainer's `.pixi/` environments.
- Or copy individual directories from `skills/` into your agent's supported skills
- location. Keep the whole directory, including references, any scripts and `LICENSE`.
- No custom installer or running service is required.
- ### Published repository (publication required)
- Use the following commands only after the matching skills and marketplace
- manifest have been published to `pymc-labs/pymc-modeling`. The published manifest
- must name both the marketplace and its plugin `pymc-modeling`. Until then, use
- the local-checkout instructions above. Local validation does not establish remote
- availability or successful remote plugin installation.
- For the Skills CLI:
- ```bash
- npx skills add pymc-labs/pymc-modeling
- ```
- For a Claude Code managed bundle:
- ```text
- /plugin marketplace add pymc-labs/pymc-modeling
- /plugin install pymc-modeling@pymc-modeling
- ```
- Choose either the plugin or the Skills CLI/manual installation, not both, to
- avoid duplicate skills.
- ## Use
- Ask your agent to perform a relevant task, or invoke the skill by name using
- your agent's skill mechanism. For example:
- - “Build a hierarchical model for these grouped observations.”
- - “Check whether these priors imply plausible outcomes.”
- - “Diagnose this posterior and assess whether LOO is reliable.”
- - “Find the shape or gradient error in this PyTensor graph.”
- - “Use pymc-extras to marginalize latent states or fit and forecast a structural time series.”
- Current modeling guidance uses the [checked stack](skills/pymc-modeling/references/pymc-extras.md#resolve-compatibility-before-choosing-an-api) and ArviZ's DataTree API.
- References retain explicitly named older-release cautions where they help existing projects. Check each reference's version scope.
- Use the consuming project's compatible Python environment.
- Optional packages support only the workflows that need them.
- Do not replace a working environment just to install these instructions.
- The [MLflow skill](skills/pymc-mlflow/SKILL.md) supports ordinary PyMC tracking
- without PyMC-Marketing. Its optional integration examples target PyMC-Marketing
- 1.2.0, which requires PyMC `>=6.3.1,<6.4.0` and ArviZ `>=1.2.0,<2.0`, plus a
- separate MLflow installation. Saving posterior artifacts, restoring native
- models and registering MMM prediction wrappers are distinct from deploying a
- service; the references make those boundaries explicit.
- The generic [MLflow tracking example](skills/pymc-mlflow/references/tracking.md) records its verified stack and compatibility limits.
- PyMC-Marketing 1.2.0's built-in MMM pyfunc prediction route forwards an unsupported
- `original_scale` argument. The [MMM reference](skills/pymc-mlflow/references/mmm-persistence.md)
- documents this reproduced upstream limitation and the working native
- restoration/prediction route; wrapper registration is not a serving guarantee.
- The [pymc-extras guide](skills/pymc-modeling/references/pymc-extras.md) routes to
- inference, exact/approximate marginalization, specialized distributions,
- state-space forecasting, prior factories and model-building tools. Its examples
- distinguish supported recovery from approximation error.
- Inspect installed APIs when adapting them to another release.
- ## Release versions
- Release tags use `vMAJOR.MINOR.PATCH`.
- `MAJOR.MINOR` identifies the primary target PyMC series.
- `PATCH` tracks revisions to this plugin, not the PyMC patch release.
- For example, `v6.4.1` can revise the skills for the PyMC 6.4 series.
- The [compatibility guide](skills/pymc-modeling/references/pymc-extras.md#resolve-compatibility-before-choosing-an-api) records exact checked versions and dependency bounds.
- Individual references retain their stated compatibility limits.
- The marketplace and plugin versions match the release tag without its `v` prefix.
- ## Maintainer checks
- These checks exercise the educational custom Op and its CPU backends, not the
- full set of modeling examples or accelerator hardware. They do not install
- anything into a consuming project's environment.
- Use Pixi 0.83.0 or newer for the platform-specific glibc declaration.
- See the [checked stack and dependency bounds](skills/pymc-modeling/references/pymc-extras.md#resolve-compatibility-before-choosing-an-api) for the extras environment and the shared PyTensor version.
- The Linux test environments require glibc 2.35 or newer. From this repository:
- ```bash
- pixi run --locked test
- pixi run --locked -e jax test -m jax
- pixi run --locked -e pytorch test -m pytorch
- pixi run --locked -e mlx test -m mlx
- ```
- For the compatible PyMC Extras stack without changing the default environment:
- ```bash
- pixi install --locked -e extras
- pixi run --locked -e extras python -c "import pymc_extras; print(pymc_extras.__version__)"
- ```
- This environment includes histogram and grouped NetCDF support for executing the
- extras reference examples. The existing pytest suite does not validate all extras
- features; runtime checks of documented examples do not establish scientific
- adequacy or accelerator compatibility.
- The default environment checks the base Op and Numba; other backend tests skip
- when their optional packages are absent. The named environments install the
- corresponding backend, with CPU-only PyTorch and MLX packages. `pixi.lock` records
- exact resolved versions, and each pytest session prints the versions actually
- exercised. Tests compare values to SciPy and weighted derivatives and curvature
- to analytic references, and check support boundaries and input errors.
- The GitHub Actions workflow runs the four custom-Op test environments with the lockfile,
- not the optional extras examples.
- Use `pixi run --locked check` to lint tests and skill scripts.
- Use `pixi run --locked format` for regression-test formatting.
- These maintainer tools are not needed to install or use the skills.
- The MLflow references contain standalone Python examples for manual tracking,
- autologging and native MMM/CLV restoration; they are not covered by the custom-Op
- pytest suite. Execute them in separate processes and disposable working
- directories with their version-matched dependencies. Their checks exercise
- artifact content, searchable metadata and restored predictions, not scientific
- adequacy. The MMM reference separates its intentional upstream-failure
- reproduction from the successful native and local-registry examples.
- ## License
- Adapted from [pymc-labs/pymc-modeling](https://github.com/pymc-labs/pymc-modeling/tree/b41e66104685ba06fce89091cf0706e9bc21872e).
- Licensed under [MIT](LICENSE). Each skill includes the same license for
- standalone distribution, including the original PyMC Labs copyright notice.
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.