mcpField
D
discernment

discernment

github.com/borkwebProductivity837 installsUpdated 1d ago
Repository

About discernment

discernment is a skill for assessing decisions and claims against evidence, especially when stakes or uncertainty warrant deeper review. It separates observations from assumptions and inference, tests for falsifiers and downside risk, and provides a conclusion, uncertainty, and next step.

Key features

  1. 01Define the decision, constraints, and uncertainty
  2. 02Separate observations, assumptions, and inference
  3. 03Check primary sources, base rates, and past outcomes
  4. 04Consider falsifiers and decision-changing conditions
  5. 05State conclusions, uncertainty, and next actions

Use cases

  1. 01Assess a high-stakes or uncertain decision
  2. 02Check the evidence behind a claim
  3. 03Record a prediction for later calibration

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/borkweb/skills --skill discernment

After installing, type/discernmentin Claude Code to confirm it loaded.

Bundled filesreferences/frameworks.md

Details

From the project README

Discernment

Assess a decision or claim against evidence. Use the full framework when requested or when stakes and uncertainty warrant it; ordinary questions do not need a judgment ceremony.

  1. Identify the decision, stakes, settled constraints and uncertainty from the brief. Ask only for missing context that changes the assessment. Do not silently substitute a different problem.
  2. Separate observations, assumptions and inference. Check relevant primary sources, base rates and actual past outcomes. If unavailable, say what remains uncertain.
  3. Look for a plausible falsifier, an underweighted downside and a condition that would change the recommendation. Apply only the lenses that help; read references/frameworks.md for a specifically requested framework. Its historical descriptions are not evidence about the user.
  4. Give the conclusion, supporting evidence, material uncertainty and one useful next action. Record a prediction and follow-up criterion when the task calls for later calibration; do not write personal memory unless requested.

Assess the reasoning, not the person's intelligence or personality. Sparse chat does not establish Dunning-Kruger, a Hogan profile or another psychological diagnosis. Do not repeat unsupported population percentages. Use a probability only with a stated basis; otherwise describe uncertainty and the check that would reduce it. A simulated premortem or dissenting perspective is a hypothesis, not observed evidence.

Comments

Markdown supported