| name | fresh-eyes |
| description | Zero-context recipient review of a finished human-facing deliverable — what is clear, what isn't, what questions they'd ask — improved with creative freedom, then repeated by a fresh-context agent until a pass changes nothing. Use after completing anything a person will read, such as a PR body, brief, docs page, dashboard, email, or essay. |
Fresh Eyes
The agent that produced an artifact cannot have fresh eyes. This pass
simulates the real reader, then verifies the simulation with an agent
who genuinely has no context.
The pass
Read the deliverable as the person who will see it on Monday morning
with groggy, tired eyes and zero context. From that seat, answer three
things:
- what is clear
- what isn't
- what questions would they ask
Then improve the deliverable however you deem fit based on those
answers. The creative freedom is real: restructure, retitle, cut,
front-load — whatever makes it land for that reader.
The repeat
Once your improvements are applied, hand the deliverable to a
fresh-context agent — one that has seen neither its production nor this
conversation — running this same skill. Repeat until a pass produces no
material change, to a ceiling of three passes; after the third, keep the
last version and say the cap was reached. Two passes is typical, and an
autonomous caller needs the ceiling to be finite.
Boundaries
- Run this before human review is requested. After handoff — a review
requested, an awaiting-human-review label, recorded evidence citing
exact text — propose edits instead of applying them.
- Improve how the facts land, never the facts: claims, numbers, and
evidence are read-only.
- The persona is the tool. Do not soften it to "review for clarity";
groggy, Monday, zero context is the standard the deliverable must
survive.
