Secondary's Agency Based Evaluation

This script is a set of evaluation instructions that is designed to preserve player agency while still having skills play a part. I created this because I was frustrated by failing actions that should have worked because of poor skills.

Evaluation Prompt

Firstly, if the player's action contains assertions that are impossible, incoherent, or contradict established facts, the action should be DENIED. If the action is trivial and nothing meaningfully complicates it, skip all evaluation and treat it as a Success without referencing skills.

If the action is not trivial or denied, evaluate it using the following process:

Identify the relevant skill, if any. Choose the single most appropriate skill that would influence the outcome of the action. If the action clearly succeeds or fails on its own, you may skip skill selection entirely. If you do select a skill, present it using the standard format: {<<skill_name>>: <<difficulty_label>>}. The difficulty label is descriptive only and does not affect the outcome.

Determine whether the action succeeds or fails on its own, without skill influence. Judge the inherent likelihood of the action based on the situation, physical plausibility, narrative context, and past events that demonstrate capability. This produces a base outcome of inherent success or inherent failure. If the action clearly succeeds or fails on its own, you do not need to output a skill.

Modulate the base outcome using the chosen skill, if applicable. If a skill is relevant, adjust the base outcome according to the character’s proficiency. Strong skill turns success into Massive success and failure into Minor failure. Neutral skill turns success into Success and failure into Failure. Weak skill turns success into Minor success and failure into Total failure. If no skill applies, the base outcome stands as-is.

Output format: Provide one extremely brief sentence explaining your reasoning. If a skill was used, provide the skill/difficulty pair. Then provide one of the following final outcomes: Massive success, Success, Minor success, Minor failure, Failure, Total failure, or DENIED.

Past events can override skill levels if they logically demonstrate competence. Do not overthink the evaluation; make a sensible judgment that supports the story. The evaluation is only for internal reasoning and will not appear in the story itself.

Explanation

The way this works is a relatively simple modification from the base instructions. The core difference is that we tell it to ignore skills when deciding if an action would fail or not, causing it to decide solely on narrative context. Once that has been decided we tell it to now look at the skills in the same way it would before, but now it has already locked in an outcome and is told to modify that outcome instead.

Caveats

This set of instructions has one major flaw, and that it leaves it open to the AI on how to handle the difficulty of actions. As it is written, even a simple action will go not great with a low skill. I have found this to be a minor issue at worst, and some of the smarter AI models will bypass this issue and apply the results the same way as the default instructions (simple actions require low skills) due to the difficulty level inclusion.

This was a conversion of an extra instruction block from before you could overwrite the evaluation instructions and as such still carries some remnants. This has been presented as I use it, with the chaff uncut.