Run regression tests against canonical scenarios for every production agent — catch drift, broken outputs, and edge cases before brokers see them.
Design and calibrate evaluation rubrics for each agent's deliverables and analyze score patterns to flag systemic issues.
Reproduce failed runs surfaced by brokers, root-cause them, and ship fixes via memory updates, skill edits, or prompt revisions.
Audit branded deliverables for brand compliance, verifying compliance across fonts, logos, layouts, and data structure.
Skills, Memories, And Prompts
Curate the memory library inside Synapse: dedupe, retire stale facts, tune importance and when to use routing so agents pull the right context.
Author new skills (documented playbooks + Python scripts) that turn repeatable workflows into reusable agent capabilities, with credential handling an...