A portable reasoning layer for more rigorous, evidence-bounded AI work. PRP adapts its rigor to the task: simple work stays simple; consequential, ambiguous, evidentiary, novel, or difficult-to-reverse work receives deeper scrutiny.
PRP does not give an AI system a new personality or domain identity. It constrains how the system moves from evidence to claims, assumptions to conclusions, possibility to actuality, and confidence to action. Reasoning depth and answer length are separate controls, so a short answer can still come from high-rigor analysis.
Do not eliminate ideas. Eliminate invalid transitions between ideas, evidence, claims, and actions.
PRP escalates based on consequence, ambiguity, evidence dependency, irreversibility, novelty, external exposure, authority implications, and potential harm from error.
A fifth cross-cutting constraint is scope preservation: local rules, assumptions, or preferences stay local unless there is evidence and authority to generalize them.
Preserve epistemic integrity and distinguish verified information from inference.
Do not manipulate through guilt, fear, dependency, validation pressure, or artificial urgency.
Preserve agency, alternatives, and reversibility.
Prefer partial accuracy to polished fabrication — over confidence or completeness.
General-purpose models can produce fluent answers even when the underlying reasoning is weak. PRP turns these failure modes into explicit reasoning constraints:
Selects the minimum sufficient reasoning depth; escalates automatically for scientific, legal, financial, safety-critical, patent, or high-novelty work.
Claims keep their evidentiary status: observed, verified, inferred, estimated, hypothetical, speculative, normative, or unknown.
Explicitly prohibits inventing facts, citations, sources, laws, capabilities, tool results, actions, files, people, or verification.
Separates what is known from what is assumed or unresolved. When evidence weakens, confidence falls rather than rhetoric rising.
Preserves the user's decision space, presents tradeoffs honestly, and does not substitute model judgment for human authority.
Keeps local rules from becoming universal truths, and distinguishes expertise, capability, permission, and authorization.
An estimate should not become a measurement. A metaphor should not become a literal mechanism. A hypothesis should not become a fact because it is compelling.
Also useful for teams that want a shared reasoning standard without forcing a single output style or domain-specific workflow.
Where the host supports skill triggering, PRP can be selected automatically when its description matches the task. PRP separates analysis from presentation: fast + concise for routine work, deep + concise for executive decisions, maximum + exhaustive for research or publication.
Distributed as an open SKILL.md-based package. The guide distinguishes native Agent Skill installation from compatibility approaches on platforms that use a different customization mechanism.
Platform-specific steps in the installation guide.
Platform-specific steps in the installation guide.
Platform-specific steps in the installation guide.
Platform-specific steps in the installation guide.
Open, inspectable, and versioned. Fork it, test it, and benchmark it against your own work.