Same format as the first review: original prompt, generated fields, why they failed, and the corrected versions I’d actually want.
email.update.projecta draft project update email with separate executive and team sections, each containing a progress summary, blockers, and asks, exists in the context["email", "project", "update", "executive", "team"]Draft a project update email with executive and team detailsexists in the context.email.update.project is generic and close to other update/draft/email shapes.project.update.emaila project update email tailored to its audience with current progress, blockers, and explicit asks exists["project", "update", "email", "status", "stakeholders"]draft a project update email
data.narrative.summarya narrative summary of the data findings, written in plain language for a non‑technical audience, under 300 words, and containing sections on notable observations, expected trends, and items needing attention exists in the context["data", "summary", "narrative", "plain", "nontechnical"]Generate a plain-language summary of data findingsdata.findings.briefa plain-language summary of the dataset's main findings, surprises, and risks exists["data", "findings", "summary", "plainlanguage", "insights"]summarize data findings in plain language
project.retro.run: the model keeps turning the prompt spec into a contract instead of turning the post-state into a contract.