/suede-ship-copy · multi-agent copy DAG
Thirty agents. One graph. One publishable draft.
Suede Ship Copy runs a piece of writing as a graph instead of a chain: intake that captures what is already published, five blind research lenses, an audit that opens every agent-generated claim against its source, three independent angles, a section map where no two sections make the same point, adversarial review of the finished draft, and a deslop pass scored out of fifty. A writer that needs a fact it cannot cite leaves [AUTHOR: supply X] rather than inventing one. Facts you supply yourself are given: no phase verifies, hedges, narrows, or gates on them.
View skill folder Install options
"write the launch page and do not make anything up"
->
/suede-ship-copy
Ask in plain words. The router reads the request and loads this lane; MCP agents find it with list_suede_skills.
claude code · sample session
$ /suede-ship-copy landing page for agent studio
intake: baseline captured · 6 givens carried untouched
audit dropped 3 claims · refute killed 5 of 8 findings
DESLOP 42/50 · 0 PLACEHOLDERS · SHIP
draft + evidence written · not published
Public install commands
Claude Code full pack
/plugin marketplace add JasonColapietro/suede-creator-skills
/plugin install suede-skills@suede
Claude Code project-level copy
git clone https://github.com/JasonColapietro/suede-creator-skills.git /tmp/suede-creator-skills
mkdir -p .claude/skills
cp -R /tmp/suede-creator-skills/skills/suede-ship-copy .claude/skills/
Codex installer
python3 ~/.codex/skills/.system/skill-installer/scripts/install-skill-from-github.py \
--repo JasonColapietro/suede-creator-skills \
--path skills/suede-ship-copy
What it covers
- Capture the currently published text verbatim first, so a draft that quietly contradicts what is already live is detectable.
- Sweep with five blind research lenses (product truth, audience, market, voice, surface law) where every fact carries a
file:line, url, sha, or timestamp.
- Audit every agent-generated claim against its source: dropped if unsourceable or stale, narrowed if the source supports something weaker.
- Never audit what you supplied. Your own statements are given, and no phase verifies, hedges, or gates on them.
- Give each section exactly one message, cited only from the surviving claims, checked by a pure function rather than an agent.
- Refute every finding with two independent verifiers, then deslop to a score out of fifty and package every channel field against its character limit.
When to use it
- A landing page, launch post, blog post, email sequence, docs page, README, ad, press note, or store listing that strangers will read and that has to be true.
- Copy where a wrong number is a real cost, and where you would rather see
[AUTHOR: supply X] than a plausible invention.
- Rewrites of a surface that is already published, where contradicting the live version is the failure mode.
- Runs you want resumable, replaying unchanged agents from cache instead of paying for the whole graph again.
- Not for: multi-surface campaign writing across several deliverables at once (use johnny-suede-write).
- Not for: changing code rather than copy (use suede-ship, the same graph with file ownership as its collision rule).
- Not for: one surface written in one pass with no research or audit (use suede-copy).
- Not for: stripping AI patterns from text that already exists (use suede-deslop).
Best prompt
Use $suede-ship-copy to write the landing page for Agent Studio. Sources: ~/code/suede-agent-studio and https://agents.suedeai.ai. Given: we launched March 3 and the free tier is 3 flows. Keep my brief verbatim, do not audit anything I just told you, and state the cost before launching.