Define an AI agent's completion criteria
Turn a task description into a concise goal with an observable result, a way to verify it and any constraints the original request depends on.
Define an AI agent's completion criteria by deciding what evidence would convince a colleague the job is finished. A request to tidy a list leaves several possibilities open. Preserving every row, assigning each one a date or marking it unresolved gives the work an observable endpoint. The goalify skill turns that description into a short condition the user can copy.
It asks how the result would be visible in the conversation. A final count, a test result or a brief excerpt may be enough. Keep constraints you actually supplied and check any suggested stopping limit. The skill produces text; starting the task is a separate action.
What it does
- State the outcome in terms someone else can verify.
- Identify evidence the working agent should show in the transcript.
- Retain restrictions that materially affect the result.
- Remove method instructions that do not determine completion.
- Add a reasonable stopping point where the task has none.
Install and use
- Install the skill in your agent and supply the original task. Drafting needs no external account; using the resulting command requires a compatible host.
- Ask for a condition that preserves all rows and reports the number with unresolved dates.
- Read the proposed wording and correct any assumptions before copying it into the intended session.
FAQ
Source
Made by Murphiyii. Original skill: goalify, published on GitHub under the Apache-2.0 licence.
View on GitHub