Brand Campaign Concept Sprint

Most campaign concepting fails the same quiet way design concepting does: the first plausible idea arrives early, looks finished, and absorbs all the refinement energy the team has. By the time someone asks whether there was a genuinely different way to answer the brief, the deck is due and the answer is no longer affordable.

Asking one agent for five campaign ideas in one prompt does not solve this. The ideas share a context window and converge into one concept wearing five headlines. And asking the same agent to critique its own favorite produces compliments with a few softening caveats, which is roughly what an internal review without a structure produces too.

This workflow separates the two jobs structurally. Each territory is developed by its own concept agent that never sees the others, and critique is a stage written into the orchestration script — against the brief, the brand guidelines, the audience insight, and the legal and claims constraints — so it happens on every run whether or not anyone remembers to ask for it.

Use the sprint at the front of any brief where the answer is genuinely open: a product launch, a rebrand announcement, a seasonal platform, a refresh of an always-on program that has gone stale. The output is a decision artifact for a creative director, not finished creative — territories with enough substance to judge, scored against the same criteria, with the strongest argument against each one attached.

Skip it when the client has already bought a direction, when the brief is really an asset request wearing a strategy hat, or when the budget only supports executing one safe idea well. The sprint earns its two to four hours when the cost of choosing the wrong territory is measured in weeks of production and a client relationship.

• Product launches where the positioning could credibly go several ways.
• Rebrand or repositioning announcements with hard legal and claims constraints.
• Always-on or seasonal programs that need a refresh, not a repaint.
• Pitches where the agency needs to show range without burning the whole creative team for a week.

This runs as a Claude Code dynamic workflow. Including the word workflow in the prompt — or running /effort ultracode — makes Claude write a JavaScript orchestration script that runs in the background: it fans out one concept agent per creative stance, waits for all of them, then fans out critique agents and a scoring pass, and only the shortlist report comes back into the main conversation. The territory drafts and the individual critiques live in script variables and on disk, not in Claude's context, which is what keeps the territories from contaminating each other.

The script can run up to 16 agents concurrently and up to 1,000 in a single run, so five territories plus a full critique matrix is comfortably inside the budget. Workflows are resumable, so a sprint interrupted after concepting can pick up at the critique stage instead of starting over. Save the working script to .claude/workflows/ in the project (or ~/.claude/workflows/ for personal use) and the sprint becomes a reusable slash command the whole team can run; the subagent definitions it calls live as markdown files in .claude/agents/.

The structural point is the same one that makes design critiques work: critique baked into the script is not optional, not deferred to whoever has time on Friday, and not softened because the room likes the person who wrote the line. Every territory faces the same reviewers with the same rubric.

Client briefs arrive as decks, emails, and meeting notes. The first human job in the sprint is to compress that into one working brief every agent receives identically: the audience and the single most useful insight about them, the business objective, the brand's voice and visual rules, the mandatory inclusions, and — critically — the legal and claims constraints, written as rules rather than vibes.

The claims section is the part agencies most often leave implicit, and it is the part that kills territories late and expensively. If the product cannot say clinically proven, or the category regulator restricts comparative claims, that belongs in the brief on day one so the critique agents can enforce it on every territory.

# Working brief: [campaign name]

## Business objective
One sentence. What changes in the business if this campaign works.

## Audience and insight
Who they are, and the one tension or truth the campaign should be built on.

## Brand rules
- Voice: [three adjectives, with one sentence each on what they exclude]
- Visual: reference brand-guidelines.md sections on logo, color, typography
- Things this brand never does: [list]

## Mandatories
- Product name and lockup usage
- Required legal line(s), verbatim
- Channels in scope for this sprint

## Claims and legal constraints
- Approved claims (verbatim, with substantiation reference)
- Forbidden claims and restricted comparisons
- Regulator or platform rules that apply (e.g. financial promotions, health claims)

## Stances (one per concept agent)
- Territory A: [creative stance, one paragraph]
- Territory B: [creative stance, one paragraph]
- Territory C: [creative stance, one paragraph]

The number of territories matters less than how different their starting positions are. A stance is a one-paragraph creative position the territory must commit to: lead with the founder story, lead with the category enemy, lead with the customer's before-and-after, lead with pure product demonstration, lead with humor at the brand's own expense.

If you cannot write three stances a reasonable creative director might each defend, the brief is probably already pointing at one answer and the sprint is overhead. Stances are also where strategy earns its seat: a planner writing five sharp stances in twenty minutes shapes the whole sprint more than any amount of prompt tuning.

• Each stance should change what the campaign says, not just how it looks.
• Include at least one stance the team finds slightly uncomfortable; safe sprints produce safe shortlists.
• Keep stances to one paragraph; the concept agent fills in the rest.

Each concept agent produces a territory package in a fixed format: a concept statement, the key message, visual direction notes, three to five sample headlines, and channel ideas for the channels named in the brief. The fixed format is what makes territories comparable in the critique stage and readable in the decision meeting.

Subagent definitions live in .claude/agents/ as markdown files with frontmatter. The repo folder for this workflow ships them under agents/ so they can live in version control; copy them into .claude/agents/ in your project to make them available.

---
name: concept-author
description: Develops one campaign territory from a working brief and an assigned creative stance. Used in concept sprint workflows; never sees other territories.
tools: Read, Write, Glob
---

You develop exactly one campaign territory.

Rules:
- Commit fully to the assigned stance, even where it creates risk. Name the risk.
- Stay inside the brief's mandatories and claims constraints; if the stance conflicts
  with a constraint, say so explicitly rather than quietly softening the idea.
- Output a territory package: concept statement (under 100 words), key message,
  visual direction notes (composition, photography or illustration direction, color
  emphasis within the brand palette), 3-5 sample headlines, and channel ideas for
  each in-scope channel.
- Write in the brand voice defined in the brief, not in agency case-study voice.
- Do not reference, imagine, or hedge against other territories.
- Write the package to the path you are given, in markdown.

The script below is a sketch of what Claude Code generates when you run the workflow prompt — it is not a library you install. The agent function runs a subagent in the background and resolves with its final answer; everything else is ordinary JavaScript holding the plan, which is exactly why the territories stay independent: the plan lives in the script, not in one model's context.

// Sketch of the orchestration script Claude Code generates for this workflow.
// agent(prompt, options) runs a subagent and resolves with its final answer.
import { readFile, writeFile } from "node:fs/promises"

const brief = await readFile("sprint/campaign-brief.md", "utf8")
const guidelines = await readFile("sprint/brand-guidelines.md", "utf8")
const stances = ["founder-story", "category-enemy", "customer-transformation", "product-proof"]

// 1. Develop territories in parallel, in isolation.
await Promise.all(
  stances.map((stance) =>
    agent(
      "Use the concept-author agent rules. Your stance is " + stance +
        ". Write your territory package to sprint/territories/" + stance + ".md.\n\nBRIEF:\n" + brief +
        "\n\nBRAND GUIDELINES:\n" + guidelines,
      { model: "opus" }
    )
  )
)

// 2. Adversarial critique: every territory is reviewed by critics that did not write it.
const lenses = ["brief-fit", "brand-and-audience", "legal-claims"]
const critiques = []
for (const target of stances) {
  for (const lens of lenses) {
    critiques.push(
      agent(
        "Use the territory-critic agent rules with the " + lens + " lens. Critique sprint/territories/" +
          target + ".md against sprint/campaign-brief.md, sprint/brand-guidelines.md, and the claims constraints. " +
          "Score each rubric criterion 1-5 with evidence. Your job is to find where this territory fails, not to balance praise.",
        { model: "sonnet" }
      )
    )
  }
}
const critiqueResults = await Promise.all(critiques)
await writeFile("sprint/critiques.md", critiqueResults.join("\n\n---\n\n"))

// 3. Scoring pass and shortlist for the creative director. No winner is picked.
const shortlist = await agent(
  "Read sprint/territories and sprint/critiques.md. Produce a shortlist report: weighted rubric scores per territory, " +
    "the strongest objection to each, any claims or brand violations (which disqualify rather than score), " +
    "blend candidates, and the open questions a creative director must decide. Do not pick a winner.",
  { model: "opus" }
)
await writeFile("sprint/shortlist.md", shortlist)

The prompt names the inputs, the stances, the agents, and the outputs, and it says workflow so Claude Code orchestrates it as a background script rather than improvising in one context window. Once a run produces a script you trust, ask Claude to save it to .claude/workflows/ so the next sprint is one command.

Run a concept sprint workflow for the campaign described in sprint/campaign-brief.md.

Inputs:
- sprint/campaign-brief.md (working brief, including claims constraints and stances)
- sprint/brand-guidelines.md (voice, logo, color, typography rules)
- The agent definitions in .claude/agents/concept-author.md and .claude/agents/territory-critic.md
- sprint/scoring-rubric.md

Orchestration:
1. For each stance in the brief, run one concept-author agent in parallel and in isolation.
   Each writes its territory package to sprint/territories/<stance>.md.
2. For each territory, run territory-critic agents with three lenses: brief fit,
   brand and audience, and legal/claims. Critics never review a territory they authored.
3. Run a scoring pass that applies sprint/scoring-rubric.md, marks any claims or brand
   violations as disqualifying, and writes sprint/shortlist.md.
4. Return only the shortlist report to the conversation. Keep territory drafts and
   individual critiques on disk.

Do not pick a winning territory. Do not soften critiques. Flag every claims risk explicitly.

Three lenses cover most agency realities. Brief fit asks whether the territory answers the business objective and the audience insight or merely entertains the team. Brand and audience asks whether the idea sounds like this brand and lands with this audience rather than with the people in the room. Legal and claims asks whether every headline and channel idea survives the constraints — and a violation disqualifies rather than scores low, because no creative director wants to fall in love with a line the lawyers will remove.

Keep the rubric to five or six weighted criteria so the shortlist meeting can hold it in one glance. Every score must cite evidence from the territory package or the brief; unsupported scores go back.

A DTC skincare brand launching a retinol alternative ran the sprint with four stances: ingredient transparency, the routine-simplification enemy, the dermatologist's-honest-take, and a customer-diary transformation story. Concepting took 35 minutes; the twelve critiques and the scoring pass took another 25.

The legal lens did the most expensive work first. The dermatologist territory leaned on clinically proven language the brand had no substantiation for, and the critique quoted the three offending headlines verbatim — a flag that would otherwise have surfaced at claims review three weeks into production. The transformation territory scored highest on audience insight but its critic showed two of its five headlines implied results timelines the brief's approved claims could not support; the package was sent back for one revision pass rather than killed.

The creative director shortlisted ingredient transparency and the revised transformation territory, took both to the client with the objections attached, and the client chose transparency — citing, usefully, the same risk the critique had named about transformation stories in a category full of them.

An agency preparing a rebrand announcement for a B2B software client ran five territories. Two died in critique on legal claims: one was built on a market-leadership claim the client could not substantiate in two of its three operating regions, and another's hero line implied a guarantee about migration timelines that contradicted the client's own contracts. Both were eliminated before any human had presented them internally, which the agency's account lead estimated saved a full review cycle with the client's legal team.

Of the surviving three, the scoring pass put the quietest territory — built on customer proof rather than manifesto language — narrowly ahead on brief fit, with the strongest objection being that it would be hard to distinguish from the category's default voice. The creative director chose it anyway and spent the recovered time on distinctive craft in the launch film, which is exactly the kind of taste call the report is designed to leave to a human.

A retail brand's always-on social program had flatlined, and the team ran the sprint to refresh it with four stances ranging from safe (polished product styling) to uncomfortable (community-sourced chaos). The scoring pass ranked the safest territory first: it fit the brand guidelines most cleanly and carried the fewest objections.

The team shipped the safe territory but, because the sprint had left the runner-up documented with its objections and sample executions, they also put the chaotic territory into a four-week paid test on a small budget. The test outperformed the safe territory on engagement rate and cost per click by margins large enough to redirect the program the following quarter.

The lesson the team kept: the sprint's scores measure fit to the brief and the brand as written, not what the audience will reward. Keeping the rejected territories on file is what made the cheap test possible.

A weak sprint produces five headlines for one idea and a critique that praises everything. A strong sprint produces territories that disagree about what the campaign should say, critiques that quote the brief and the offending lines, and a shortlist a creative director can defend to a client a month later.

The sprint can generate genuinely different territories and argue about them honestly when the structure forces it. It cannot know which risk the client will tolerate, whether a line is funny or merely shaped like a joke, or what an audience will actually reward — the social refresh case is the standing reminder that the highest-scoring territory is the best fit to the brief, not a prediction of performance.

Claims approval also remains a human and legal responsibility. The legal lens catches violations of the rules written into the brief; it cannot catch rules nobody wrote down, and it does not replace the client's legal review. Treat its flags as an early warning system, not a sign-off.

• It cannot replace creative judgment or brand taste; it makes the options and objections explicit.
• It cannot predict in-market performance; testing decides that.
• It cannot enforce constraints the brief never stated.
• Scores are comparable within one sprint, not across briefs or clients.

Save the working script to .claude/workflows/ and the agent definitions to .claude/agents/, and the sprint becomes a command the studio runs at the front of every open brief — with the brief template, rubric, and critique lenses improving a little each time.

1. Compress the client brief into one working brief: objective, audience insight, brand rules, mandatories, claims constraints.
2. Write 3-5 creative stances a reasonable creative director could each defend; include one uncomfortable one.
3. Run as a dynamic workflow: one concept-author agent per stance, developed in parallel and in isolation.
4. Run territory-critic agents over every territory with three lenses: brief fit, brand and audience, legal and claims.
5. Score against the rubric; claims and brand violations disqualify rather than score.
6. Generate a shortlist report: scores, strongest objections, disqualifications with quotes, blend candidates, open questions.
7. The creative director picks, blends, or sends a territory back for one revision pass.
8. Record the decision and keep the rejected territories on file for testing and the next brief.