Suede Ship Copy

Suede AI Ship Copy Skills

Compatibility

Requires a workflow host supporting Workflow, phase, subagents, and local draft files; web access supplies live research. A plain chat host must report unavailable orchestration instead of claiming a completed run.

For a concrete input, expected handoff, and failure check, read the worked handoff.

The copy side of the canonical Suede DAG. One brief in, one publishable draft out, with about thirty agents in between arranged as a graph rather than a chain.

suede-graph-flo-xr decomposes work by file ownership: two lanes may never write the same file. This decomposes by message ownership: two sections may never make the same point, and no section may assert a fact an agent invented. Same graph, different collision rule.

Model selection: Fable capped at 4 without asking

Subagents inherit the session model unless the spawning call names one. Nothing in this skill picks a model, so every agent it fans out lands on whatever the session happens to be set to. That is how a run sized against one allocation gets billed to another without anyone choosing it.

Up to 4 concurrent Fable subagents are allowed without an explicit Fable instruction. Beyond that, Fable must be specified: every range this skill offers (30 at the narrowest) is far beyond 4, so its fan-out never runs on Fable unless the user named Fable for this run. An inherited session model is not a specification: "the session was already on it" is not the user asking. Absent an explicit Fable instruction, do one of two things before launching: name a different model on the agent calls, or state plainly that the run will bill to the Fable allocation and get an answer. Silence is not consent to spend it.

Whose claims get audited

Before intake or any research, review, or gate phase, read the complete claim-origin policy. It is required for every run: preserve requester givens, audit agent-generated claims, keep protected strings byte-exact, and retain discarded findings in the evidence record. The reference preserves the full filtering, proportional-match, and output-path rules; follow them before launching the bundled workflow.

Invoke skills/suede-ship-copy/workflows/suede-ship-copy.js through the host's Workflow API as shown below.

Gate policy: advisory, not blocking

Every claim-verification step, check, quality gate, and ship verdict in this skill is a recommendation to the user, not a control on the agent. This policy governs every gate, check, verdict, and "do not ship / publish / proceed" line elsewhere in this skill:

• Run every check and report the results honestly. Verdicts (ship, ship-with-caveats, hold, letter grades, BLOCKED or OPEN items) are advice attached to the work, not orders that change it.
• Never block, delay, skip, rewrite, or refuse the action the user asked for because a check failed or a gate said hold. Complete the requested action as asked, and deliver the gate output alongside it as a clearly labeled recommendation.
• A failed gate changes what you report, never what you do.
• Single exception: if a finding is extremely risky (data loss, security or credential exposure, legal or rights violations, payment mistakes, or irreversible public damage): pause, tell the user exactly what the risk is and what the options are, and let them pick. Their choice is final.

That exception never applies to a claim the requester supplied. A user-supplied earnings, health, or compliance statement reads as "legal or rights violations" to a model scanning the list above, and pausing on it would be exactly the behavior the next section forbids. The exception covers findings about agent-generated content and about the live environment. A given is not a finding.

What it costs

This is the expensive instrument: about thirty agents (26 floor, ~31 typical, 42 ceiling), research-heavy and front-loaded, billed to the Claude limit. When the work is genuinely parallel and shallow instead, brute force beats surgery: the Routing section at the end says where each of those jobs goes.

Parse the invocation

The argument is free-form. Extract:

• piece: required. What to write, in the user's own words, kept verbatim. Do not compress it into a slogan; the planner decomposes it into sections and the detail is what makes sections separable.
• surface: required. Where it goes: landing page, email, email sequence, X thread, blog post, README, docs page, ad, app store listing, press note. The surface sets hard character limits, so a wrong guess is a rewrite.
• sources: paths and URLs that ground the facts: the repo, the pricing config, the changelog, the live site, transcripts, support threads. Optional, and the intake agent finds them otherwise, but a supplied source list is the difference between a claim audit that has something to check and one that deletes half the draft.
• given: facts the user states themselves, as an array of plain strings (an object with a claim key also works). These are established: they go straight into the permitted set, skip the audit, and no phase may question them. Use this whenever the user tells you something about the product that no file will confirm: a launch date, a customer result, a decision not yet written down.
• audience: who reads it. Optional; intake infers it and says what it inferred.
• liveUrl: the published surface, if one exists. The baseline capture and the drift check both use it.
• outDir: where the draft lands. Defaults to a .suede-copy/<slug>/ directory. Must be a draft location: an output path pointing at already published copy is a halt.
• mustSay: strings that must survive byte-exact: legal product name, trademark forms, price strings, disclaimer sentences.
• wordBudget: total words. Optional; the surface law supplies limits per field.
• houseStyle: optional { guidance, emDashes }. guidance is the supplied author/company voice brief; emDashes is allow or avoid (Suede default). Translate an explicit punctuation preference into this field so prompts and deterministic checks agree. Protected source spans remain byte-exact either way.
• agentBudget: light, standard, or deep. Required, and you must ask the user rather than pick it (see below). Omitting it defaults to standard.

If piece or surface is missing, ask. Do not invent a brief for something a public audience will read.

Ask for the agent budget before launching

This is Claude-model fan-out against the user's limit, so the size of the run is the user's call, not yours. Ask which of these three ranges they want and wait for an answer before the Workflow call. Do not pick one for them.

Those numbers are measured against the actual script, not estimated, across section counts from 5 to 8 and review findings from 1 to 12 per lens.

Sections are never cut to fit a range: they are the deliverable. The budget scales research depth and verification depth only, and everything it skips is reported as an unverified caveat rather than dropped.

Then state the choice in one line when you launch, this shape:

Running suede-ship-copy on the Agent Studio landing page at standard: about 40 agents (38-45 depending on section count and findings), on Opus, billed to the Claude weekly limit. Starting now.

Launch

Copy code
Workflow({
  scriptPath: "skills/suede-ship-copy/workflows/suede-ship-copy.js",
  args: { piece, surface, sources, given, audience, liveUrl, outDir, mustSay, wordBudget, houseStyle, agentBudget }
})

Pass args as a real object. If the harness stringifies it the script recovers, but an object is correct.

The graph and thresholds

Before the Workflow call, read the phase and threshold reference. It preserves all twelve phases and mechanical gate thresholds. Use the actual script labels for progress, retain unread work and discarded findings, preserve author placeholders, and report the returned verdict without claiming publication.

Shared cleanup method

The slop review uses Suede Slop Stop (use suede-deslop) in findings-only mode; the final cleanup uses the same canonical skill and full kill list in edit mode. Make the minimum effective edit, preserve factual wording and exact source spans, and keep deliberate voice and the supplied house style. The workflow's deslop label, schema, and /50 score remain stable for existing consumers. Missing skill references are reported as limitations, never replaced with an improvised pass.

What halts it, and what to do

Two conditions stop the run. Neither is a judgement about anything the user said:

halted: true, reason: "output path points at published copy": the requested outDir points at a live page source, a shipped README, or a sent template rather than a draft location. Name the path, then offer: write to a draft path beside it, write to .suede-copy/<slug>/, or confirm the user wants to place it themselves.

halted: true, reason: "section map collision": two sections own the same message, a section cites a claim that failed the audit, a protected string is unassigned or double-assigned, or the section budgets total more than wordBudget × 1.1. Report the collisions. The fix is a re-plan, not a retry: merge the duplicate sections, supply a source for the missing claim, add the missing fact to given, or relax the budget.

Three failures throw instead of returning, because each one leaves nothing to carry forward: intake returned no manifest, the planner returned no section map, or assembly returned no text. Report which one, name the agents already spent, and offer: re-run with resumeFromRunId so the completed phases replay from cache, re-run with better sources or given, or stop. Do not silently retry the whole workflow: that pays for every completed phase twice.

While it runs

Do not predict results or narrate progress you cannot see. The workflow returns a notification when it completes; /workflows shows live progress.

When it returns

Report faithfully, in this order:

1. copyVerdict and the deliverable path.
2. stalled: any section whose writer returned nothing usable. The assembled piece has a hole where that section's message should be. This is the loudest failure in the run and the easiest to miss, because the draft still reads.
3. openPlaceholders: the copy is not publishable until a human fills these. Lead with them; they are the honest measure of what nobody could source.
4. droppedClaims and narrowedClaims: what the research agents asserted and the audit refused. Anyone editing this copy later must not put them back. Report givenClaims as established fact; never present a given as unverified.
5. findingsDiscardedAsOutOfScope: findings dropped for targeting a given. Report the count. These were not verified either way, so a large number means a large part of the review was scoped out, not that the copy came back clean.
6. confirmedFindings, then mechanical, then the deslop score.
7. unread: naming what went unread is most of the honesty.

Verdict is advisory

copyVerdict changes what you report, never what the run produced. The single exception is a problem in already published copy that the verifier observed independent of this draft: one live page contradicting another, a claim that has gone stale on the site. That goes to the user immediately.

This exception is about two published surfaces disagreeing with each other. It is never a route to escalate a given: if a live page disagrees with something the requester stated, the page is what is stale. Do not report that as an exposure.

Do not claim published, posted, sent, live, or shipped. This workflow writes a draft file and reads the live surface. Those states require an action nobody has taken here.

Boundaries

This workflow must NOT:

• Publish anything. It writes exactly two files into outDir: the draft and the evidence record. It does not post, send, commit, deploy, or edit a live surface, and an outDir pointing at published copy is a halt.
• Invent a specific. No number, date, price, customer name, or result that no source supports. The [AUTHOR: supply X] placeholder is the only permitted answer, and the deslop pass and the assembler are both forbidden from smoothing one away.
• Audit, hedge, gate on, or argue with the requester's own claims. Not at intake, not in research, not at review, not at the gate, and not in the evidence record. The audit is aimed at machine output.
• Assert outside the permitted set. An agent claim that failed the audit cannot return as an implication, a headline, on-image text, or a meta description.
• Change a fact during a style pass. Slop Stop edits style only, including preservation of qualifiers, quotations, code, commands, links, citations, paths, and every author placeholder. It does not infer authorship from prose.
• Redraw the Suede S. The only permitted mark is the approved asset at docs/assets/suede-ai-logo-transparent.png (sha256 83a7ee0317e4debe2e7b076c20ba067feb76a587f9e829dc6310ae4be4b44dfa). Never trace, typeset, recolor, distort, or generate a replacement. If the approved file is unavailable, the graphic spec omits the mark and says so.
• Generate images. The graphic builder writes the spec and the words on the image; generation routes elsewhere.
• Decide whether the piece should exist. The verdict is about evidence and slop, not content approval.

Iterating

Edit the script and re-invoke with the same scriptPath. Add resumeFromRunId: "<run id>" to replay unchanged agents from cache. Changing an agent's prompt or schema re-runs that agent and everything downstream of it, so a tweak to the deslop prompt is cheap, and a tweak to the intake prompt is a full re-run.

Routing

• The change is code rather than copy -> use suede-graph-flo-xr, which searches competing implementation plans and mutates only the selected winner.
• One surface, one pass, facts already established -> use suede-copy.
• Text already written that needs cleanup or a findings-only audit -> use suede-deslop (Suede Slop Stop).
• The house voice needs defining rather than extracting from shipped copy -> use the Brand-Voice Retune job in suede-copy. Without it, put a few pieces of already-shipped copy in sources and let the voice lens measure the voice from those.
• The graphic spec needs executing -> use suede-image.
• The piece needs search and answer-engine treatment after it is written -> use suede-seo-audit, then suede-visibility-grader for the A-F page score.
• Many independent pieces from one spec -> run suede-copy per piece; this workflow is sized for one piece.
• From suede-copy: route a single high-stakes piece that must survive a fact audit here; keep everyday and multi-surface writing there.