One-click install
npx skills add https://github.com/goodfire-ai/causalab --skill interpret-experiment
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: interpret-experiment
Source: https://github.com/goodfire-ai/causalab/tree/main/.claude/skills/interpret-experiment
Command: npx skills add https://github.com/goodfire-ai/causalab --skill interpret-experiment

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It prevents research results from being hard to summarize by automatically converting an experiment plan plus produced artifacts into a single, evidence-grounded report.

Core Features & Use Cases

  • Plan-to-evidence scoring: Extracts objective, success criteria, and hypotheses from the session plan, then matches them to shipped and session-local analysis outputs.
  • Artifact-aware interpretation: Discovers what actually ran from resolved configs and on-disk artifacts, including session-local methods and analyses, and records plan-vs-reality gaps.
  • Optional paper comparison: When replication context is present, compares key metrics against the paper’s reported findings and annotates discrepancy categories.
  • Figure embedding and structured output: Copies/symlinks curated figures into the session result folder and writes a consistent result/REPORT.md, while logging issues separately.

Quick Start

Run the experiment with /run-experiment, then invoke interpret-experiment (or let it auto-run) to generate ${SESSION_DIR}/result/REPORT.md from the plan and artifacts.

Frequently Asked Questions about interpret-experiment

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I generate an automated analysis report from experiment runs?

You can generate an automated analysis report by matching a planned objective, success criteria, and hypotheses to produced artifacts from a completed experiment session. It reads session markers, plan files, and resolved runner configs to create a consolidated result.

What is mechanistic interpretability experiment interpretation?

Mechanistic interpretability experiment interpretation is the process of converting a causal abstraction research plan plus produced artifacts into a single, evidence-grounded report that scores hypotheses against actual analysis outputs.

How do I compare my research results against a replication paper?

You can compare results against a replication paper by providing optional paper context during experiment interpretation. The tool compares key metrics against the paper's reported findings and annotates discrepancy categories in the final report.

Can I embed figures into research reporting outputs automatically?

Yes, automated research reporting can embed figures by copying or symlinking curated figures into the session result folder. It then writes a consistent REPORT.md file with the figures embedded directly.

What happens when my experiment plan does not match produced artifacts?

When an experiment plan does not match produced artifacts, the interpretation process records plan-vs-reality gaps. It discovers what actually ran from on-disk artifacts and logs these issues separately from the main report.

Do I need to use run-experiment before generating an experiment report?

Yes, you need a completed run-experiment session before generating an experiment report. The reporting tool applies to completed sessions and reads session markers, plan files, and resolved runner configurations to function correctly.