AgentSkillsCN

fpf-problem-framing

创建FPF异常记录（问题界定 + 多重假设 + 预测 + 测试计划）。在调试、调查、需求模糊，或面对“为什么会这样？”这类问题时，必须予以调用。

SKILL.md
--- frontmatter
name: fpf-problem-framing
description: Create an FPF Anomaly Record (problem framing + multiple hypotheses + predictions + test plan). MUST be invoked for debugging, investigation, unclear requirements, or any "why is this happening?" task.
argument-hint: "[topic-or-short-title]"

Goal

Create a written Anomaly Record under .fpf/anomalies/ that captures:

  • •the anomaly/problem (framed precisely),
  • •run-time observations (facts),
  • •constraints/invariants,
  • •multiple hypotheses (abduction),
  • •predictions (deduction),
  • •a concrete test/evidence plan (induction).

This is "thinking through writing".

Output

Create a new file:

  • •.fpf/anomalies/ANOM-${CLAUDE_SESSION_ID}--<slug>.md

Use the template:

  • •.fpf/templates/anomaly-record.md

If the user passed arguments, use them to derive <slug> and the title. If not, derive a short slug from the current task.

Procedure (ADI)

  1. •Abduction
  • •Write the anomaly in one crisp paragraph.
  • •List at least 3 plausible hypotheses.
  • •For each hypothesis, add a plausibility lens:
    • •parsimony,
    • •explanatory power,
    • •consistency with known facts,
    • •falsifiability.
  1. •Deduction
  • •Select a prime candidate hypothesis (still provisional).
  • •Derive predictions:
    • •what should be observed if true?
    • •what should not be observed?
    • •what would falsify it?
  • •Note dependencies and implications.
  1. •Induction
  • •Define the minimum viable tests to discriminate hypotheses.
  • •Prefer cheap tests first.
  • •Link to where evidence will be recorded (.fpf/evidence/...).

Quality bar

  • •Separate run-time observations (facts) from design-time assumptions.
  • •If you are uncertain, say so explicitly and propose how to reduce uncertainty.
  • •Keep the record actionable: it should tell a future reader what to do next.