Generate causalab task packages from markdown specifications with validation.
npx skills add https://github.com/goodfire-ai/causalab --skill setup-task
Please help me install this Agent Skill. Skill: setup-task Source: https://github.com/goodfire-ai/causalab/tree/main/.claude/skills/setup-task Command: npx skills add https://github.com/goodfire-ai/causalab --skill setup-task
The setup-task Skill turns a task specification into a complete, convention-compliant causalab/tasks/<task_name>/ package so you can immediately run causal abstraction experiments instead of wiring files by hand.
config.py, templates.py, causal model code, counterfactual generators, token position logic, checker, metrics, and a summary notebook.Run the skill by providing the path to your markdown task specification, for example: Invoke /setup-task path/to/task_spec.md and then confirm the extracted fields when prompted.
High-intent search queries and answers about installing and using this skill.
To create a mechanistic interpretability task package from a markdown specification, provide the MD file path to generate a complete folder with config, templates, causal models, and counterfactuals automatically. This ensures all generated files comply with causal task conventions.
Causal abstraction task setup is the process of defining causal models, counterfactuals, and token alignment checks for interpretability experiments. You need it when exploring or investigating model behaviors to ensure inputs and outputs are consistent with the chosen model.
To validate token alignment and output checking for a causal task, the setup process loads the specified model and runs end-to-end validation. This verifies output token modes, variable naming conventions, and accuracy across the generated task modules.
Yes, you need an active research session to generate counterfactuals for interpretability experiments. The task generation process requires an active session to properly wire causal models, enforce conventions, and run end-to-end model validation checks.
Yes, you can use a PDF to create a causal task specification if you lack a markdown file. The generation process supports interactive or PDF-driven spec creation with approval checkpoints to extract the necessary task fields.
The best way to ensure variable naming conventions are enforced in causal model code is to generate the task modules from approved templates. This automatically enforces output token modes and standardizes variable names across the package.
tianhao909Inspect and manipulate transformer internal activations and weights for mechanistic interpretability.
leonvanzylCreate, evaluate, and iterate Claude skills with automated test prompts.
cesaregarzaDefine ability families, AP rungs, and token constraints for mechanistic interpretability experiments.
mlc-aiCreate MoE model skeletons for PithTrain and wire them into training and inference workflows.