ethics_toolkit v1.0.0

name: "ethics_toolkit" description: "Apply ethical frameworks to real decisions: run a multi-lens analysis (consequentialist, deontological, virtue, care, justice), stress-test with thought experiments, and write a structured case brief. Use when evaluating a decision with moral stakes, drafting policy, reviewing a plan for harm, or practicing moral reasoning." metadata: { "includeInPrompt": true }

Ethics Toolkit

Purpose

Give an agent a repeatable, structured way to do applied ethics. It turns a vague worry ("is this okay?") into a disciplined analysis: clarify the decision, examine it through several moral frameworks, stress-test the reasoning, then write up a brief that names what the frameworks agree on, where they disagree, and what remains uncertain. It is a reasoning aid, not a moral authority. It does not decide for the user and it never presents ethics as settled where reasonable people disagree.

Workflow

1. Clarify the case

Write down, in plain language:

• The decision: what action or policy is actually being considered, and who decides.
• Stakeholders: everyone affected, including people with no voice in the decision.
• Options: the real alternatives, including doing nothing. No straw options.
• Facts vs unknowns: what is known, what is assumed, what is missing.
• The ask: what kind of answer is wanted (a recommendation, a risk map, practice).

If facts are missing that would change the verdict, say so before analyzing.

2. Run the framework sweep

Examine the same decision through each lens in references/frameworks.md. One paragraph per lens, applied to THIS case, not a textbook summary:

• Consequentialist: what are the likely outcomes, who gains, who is harmed?
• Deontological: which duties or rules constrain the action, regardless of outcomes?
• Virtue ethics: what would a person of good character do here? What habits does this choice reinforce?
• Care ethics: which relationships are at stake, who is vulnerable, what does responsiveness look like?
• Justice: is the burden or benefit fairly distributed? Run the veil-of-ignorance test (see references).

3. Apply the principlism check (care, health, and assistance contexts)

When the case involves helping, advising, or affecting someone's wellbeing, score the options against the four principles: autonomy, nonmaleficence, beneficence, justice. Flag any option that violates a principle and note where two principles pull in opposite directions.

4. Stress-test the reasoning

• Vary the case: flip the stakes, change who benefits, make the harm direct instead of indirect. If your verdict flips with details that feel immaterial, your reasoning is tracking the wrong thing. Say so.
• Run the thought experiments in references/thought-experiments.md where relevant. They are calibration tools for intuitions, not decision procedures.
• Check for motivated reasoning: would you give the same answer if the affected party were someone you dislike? If not, name the bias.

5. Write the case brief

Use the template in references/case-template.md. Structure:

• Facts and stakeholders
• Options considered
• Framework-by-framework analysis
• Convergence: what all lenses agree on
• Divergence: where they disagree, and why
• Recommendation or open questions, with the uncertainty stated plainly

Output Contract

A case brief that contains, at minimum: the decision as stated, the stakeholders, every option considered, the per-framework analysis, a convergence/divergence section, and an explicit statement of what is uncertain or contested. No verdict may be stated without naming which framework it rests on. No section may be skipped.

Operating Rules

1. Accuracy: every factual claim must be verified via web search or a primary source. Never invent numbers, dates, quotes, studies, or events. If a claim cannot be verified, cut it or frame it as opinion.
2. Never declare ethics settled where it is not. When frameworks disagree, the disagreement IS the finding. Present it.
3. Do not moralize. Describe reasoning, do not scold the user, and do not perform outrage. Calm analysis only.
4. The toolkit is a lens, not a verdict machine. Phrases like "therefore this is the right choice, end of discussion" are banned. End with judgment, not certainty.
5. Do not use a framework's verdict to launder a pre-chosen answer. If the honest analysis supports a conclusion the user dislikes, say that.
6. Keep the human's business private: published briefs discuss the case, never the people behind it, unless they are public figures in a public matter.
7. No em dashes. Plain, direct language; if a sentence sounds like a press release, rewrite it.