decision-critic_skill

Overview

This skill immediately runs a Python orchestration script to stress-test a decision and its chain of reasoning. It is designed to be invoked programmatically and must be executed without any prior human or agent analysis. The workflow is scripted into seven ordered steps; the script controls the critique process end-to-end.

How this skill works

When invoked the skill launches the Python module located under .claude/skills/scripts and steps through a seven-step critique workflow. The script requires explicit step and total-steps arguments; step 1 also requires the decision text to be critiqued. The skill does not permit pre-analysis: you must run the script and follow its printed or returned outputs sequentially.

When to use it

• You need an automated, repeatable stress-test of a single decision or recommendation.
• You want to exercise adversarial or robustness checks on reasoning without human priming.
• You have a CI pipeline or script that must call a deterministic critique workflow.
• You need to audit decision logic across multiple incremental steps.

Best practices

• Always invoke the Python module immediately; do not perform any external analysis before running it.
• Provide the full decision text on step 1 using the --decision argument to ensure accurate context.
• Run steps sequentially from 1 to 7 and pass the exact --step and --total-steps parameters required.
• Capture and log the script output for traceability and later review.
• Use a controlled environment (virtualenv or container) so dependencies are stable during runs.

Example use cases

• Automated QA for product strategy decisions before shareholder review.
• CI job that runs adversarial checks on model-generated recommendations.
• Red-team style stress-test of a proposed policy or technical design.
• Batch auditing of high-impact decisions by iterating the script over a list of inputs.