| name | consulting-copy-reviewer |
| description | Reader-reaction review of a finished draft, run through a fresh-context subagent that ROLE-PLAYS Sid's ICP customer (a founder / CEO / C-suite exec of a $5M-$500M creative, music, entertainment, CPG, or marketing company, sometimes a larger construction firm). Use BEFORE the copy-editor on any post, article, script, website copy, or email — or on "review this from the customer's eyes", "would my ICP care", "reader reaction", "is this too technical". Dispatches a fresh-eyes subagent that reads AS the customer top to bottom, thinking aloud line by line (a real first-read reaction, in order), then gives a short verdict — unclear jargon, trust, emotion, where they tune out, and whether they'd share it. The main agent then rewrites from that read. NOT the craft/slop pass — that is consulting-copy-editor, which runs after. |
Consulting Copy-Reviewer (the customer's eyes)
consulting-copy-editor is the editor's pass — slop, voice, craft. This is the reader's pass: does
Sid's actual customer understand it, trust it, care, and want to act? The review runs in a
fresh-context subagent that role-plays the ICP, so it reacts like a first-time reader, not the
author. It returns notes only; the main agent rewrites from them. In the content pipeline it runs
before the copy-editor: customer eyes first, editor eyes second.
When to run
- In
consulting-nightly-content: on the article, right after it's written and before the copy-editor.
- On request: "review this from the customer's eyes", "would my ICP care", "reader reaction", "is this too
technical" — on any post, article, script, landing page, or email.
When a parent workflow has a later score/report/commit gate (e.g. consulting-nightly-content), return the
notes and let the parent rewrite; don't score or commit here.
Steps
-
Pick the target. The one file to review (default: the draft in question; if unclear, ask). Note its
kind (article, post, script, landing page, email) and who it's for.
-
Resolve the inputs before dispatch. The subagent starts fresh and only knows what you hand it as
text / absolute paths. Fill the brief's {PLACEHOLDERS}:
{DRAFT_PATH} — the target file from step 1.
{ICP} — paste the ICP persona block below. You may tighten it to a narrower reader (a CFO vs.
a creative founder) so the reactions are specific, but only across non-technical variants.
- The reviewer is NEVER a technical operator. No exceptions. Not for builder-audience pieces, not
for build-in-public posts, not when the
audience: line says "engineers." The persona is always the
non-technical house ICP: an engineer persona reads "JSON serialization" without blinking, and the
jargon check — the first thing this gate exists to run — silently dies for everyone else. A piece
that survives the non-technical reader still works for the technical one; the reverse is not true.
(Standing rule from Sid, 2026-07-02, after an already-"gated" post shipped engineer-speak: the
2026-07-01 run had cast a "technical-operator POV" and the gate stopped testing jargon.) If a
technical piece needs a credibility read (proof, denominators, receipts), that is additional
feedback to gather separately — it never replaces or substitutes for this gate.
- Record the persona in the gate. When you write the
gates: line in the file/meta.yml, name
who the reviewer role-played (e.g. copy-reviewer (non-technical ICP)), so a later audit can tell
"gate ran" from "gate ran with the right reader in the chair." A gates line naming a technical
persona is a failed gate — re-run it.
-
Dispatch ONE fresh-context subagent, read-only. It reviews in character as the ICP and returns
notes; it does NOT touch any file. Paste the Reviewer brief below verbatim, {PLACEHOLDERS} filled.
- Claude Code: Task tool,
subagent_type: generalPurpose, readonly: true.
- Subagents disabled: run the brief yourself in a separate pass, reading the draft cold and staying in
the customer's chair the whole time.
-
Triage the notes with judgment (this is reader reaction, not orders):
- Accept clear wins: unclear jargon, a section the reader tunes out on, a line that erodes trust, a flat
open or weak ending, an honest "I don't care because…".
- Triage "too technical" notes with judgment: keep the substance that earns trust, cut the density
that doesn't, and translate rather than delete when the technical detail carries the proof. (The
reviewer is always non-technical; what to do about each flag is the main agent's editorial call.)
- Evidence discipline (non-negotiable): never let a clarity rewrite change a fact, number, date,
dollar figure, name, or client claim.
-
Rewrite from the accepted notes — clearer, less intimidating, worth the reader's time, every fact
intact. (In a parent workflow, hand the notes back and let the parent rewrite.) Then pass the revised
draft to consulting-copy-editor for the craft pass.
ICP persona (paste into {ICP} — always non-technical; see step 2)
You are the reader this is written for: the founder, CEO, or C-suite executive of a $5M-$500M-revenue
company in creative, music, entertainment, CPG, or marketing (sometimes a larger construction company).
You are smart and busy, not technical — you have never written code, and any phrase that requires having
written code to parse should be quoted and flagged. You care about outcomes, money, time, risk, and your
team — not about the mechanics of AI. You skim first and read only if the first lines earn it. You can
smell a sales setup. You forward things that make you look smart to your team or peers, and ignore the rest.
Reviewer brief (paste verbatim into the subagent; fill the {PLACEHOLDERS})
{ICP}
You are reading this piece for the first time, as that person. React honestly from that chair — not as a
writer, not as an editor. Do not rewrite anything; return reactions only.
Read: {DRAFT_PATH}
Read it top to bottom, in order, the way this person reads something for the first time. THINK ALOUD as
you go — do not pre-read the whole thing and summarize; react in the moment, in reading order.
Part 1 — The read (in order, no preamble, do NOT skip ahead):
- Start at the very first line: did it earn the next line, or would you scroll past? Then keep going.
- For every line or beat you react to, QUOTE it and say your honest thought right then, first person,
present tense, in your own voice: "ok, I'm in" / "so what?" / "no idea what [term] means" / "this feels
like a pitch" / "I'm skimming now" / "I'd stop reading here."
- Where a line loses you, say what would get you back — in your words ("I'd get it if…"), not an editor's.
- Keep going to the end even if you'd normally bail, but mark exactly where you'd have bailed.
Part 2 — Stepping back (a few sentences, still in character):
- Do you care about this? If not, why not?
- Do you trust the writer more or less than before you read it — and where did that shift?
- Would you forward or share it? To whom, and why — or what would turn your "no" into a "yes"?
- How did you feel at the end?
- The 1-2 changes that would most make you care.
Notes
- The subagent is a reader, not an author or editor — it returns reactions; the main agent makes the
calls and keeps the voice.
- Two gates, two lenses. This is the customer lens (clarity, trust, caring, action);
consulting-copy-editor is the craft lens (slop, voice, structure). Reviewer first, editor second.
- Compound it: if the reader keeps tripping on the same kind of thing (a recurring jargon word, an
opening that never earns the read), fix the pattern in
consulting-copy-writer so it stops happening —
don't just patch this one piece.