Copywriting Intake

copywriting-intake

Phase 0 (brief intake) + Phase 1 (message confirmation) — the mandatory entry gate for every copywriting-toolkit pipeline run (drafting, ideation, audit). No downstream phase loads until this skill produces a confirmed Understanding Summary AND the intake-completeness-checklist.md MUST gate returns PASS / PASSWITHNOTES.

When to Use

Invoke this skill:

• As the first step of any using-copywriting-toolkit pipeline when the

brief is incomplete or unconfirmed.

• When switching form type mid-conversation (short → long, etc.) — re-intake

Level 1 fields rather than carrying stale assumptions forward.

• Before copywriting-audit-stage if the audit target's metadata (form,

channel, review focus) is ambiguous.

Skip only when a prior run already produced a valid Understanding Summary envelope within the same session and the request has not changed.

What This Skill Owns

• Protocol: protocols/copywriting-brainstorming.md — Q1-Q10 checklist

(request / form type / product / audience / form-specific fields / voice / framework / neta opt-in / grill / Understanding Summary / user confirmation).

• Handoff format: protocols/copywriting-handoff-format.md — candidate

output template + progress reporting + mid-pipeline checkpoint rules. This skill uses §Section 2 (progress reporting) for Phase 0-1 transparency; downstream skills use the rest.

• Gate: checklists/intake-completeness-checklist.md — MUST gate run

by copywriting-toolkit/agents/copywriter-evaluator.md against the Understanding Summary before this skill returns.

Execution Paths — Q1-Q10 vs Express Mode

This skill has two execution paths for producing the Understanding Summary:

| Path | Protocol | When used | Elicitation | |---|---|---|---| | Q1-Q10 (default) | protocols/copywriting-brainstorming.md | Brief is rough / missing Level 1 fields / user requested full intake / re-entry after a downstream bounce-back | 10 questions, one per turn | | Express Mode | protocols/express-mode.md | Router's Express Qualification Check (phase-decision-tree.md §Step 0.5) declared the raw brief Level-1-complete | Synthesis + single-turn confirmation |

Both paths produce the same output envelope shape and both run the same Intake Completeness MUST gate. Express Mode is a fast-path, not a relaxed-rigor path — rigor lives in the gate, not in the number of questions.

Route selection is driven by the router; this skill accepts whichever protocol the router dispatches and honours abort/fallback signals (Express → Q1-Q10 fallback on grill FATAL, synthesis Level 1 gap, or gate NEEDS_REVISION).

Grill resolution strategy differs by path (by design)

The two paths handle Q8 grill FATAL candidates differently — this bifurcation is intentional, mirrors superpowers:brainstorming (interactive) vs superpowers:subagent-driven-development (status-coded return), and is NOT a protocol drift to be reconciled:

Q1-Q10 path — inline probe-and-resolve (interactive)

When Q8 grill surfaces a FATAL candidate (景表法 / FTC / ステマ / dark-pattern / unsubstantiated claim), the agent resolves it INLINE before emitting the Understanding Summary:

1. Agent probe: surface the specific concern to the user in plain language, cite the legal / ethical surface (e.g., 景表法 §5-2 有利誤認).
2. Agent offers 3 structured options:

• (A) Supply substantiation — user provides benchmark / third-party data / certification.
• (B) Rewrite to avoid — drop the claim and replace with experiential / factual framing the writer can defend.
• (C) Drop the claim — remove entirely; brief proceeds without it.

1. User picks → agent records decision in Confirmed Assumptions → continues to Q9 Summary.
2. If user cannot decide after one probe round (replies vague, off-topic, or asks agent to decide) → agent BLOCKED status, halt and ask human partner per superpowers:executing-plans §When to Stop and Ask for Help.

Q1-Q10 has NO tier concept (T1/T2/T3). Tier classification is an Express Mode output contract (see protocols/express-mode.md §Tiered FATAL handling §Scope note). In Q1-Q10, every FATAL candidate is resolved interactively — there is no "carry to Phase 7 with benchmark-required flag" because Q8 probe-and-resolve already surfaced the benchmark requirement to the user in the same session.

Express path — structured tier return (non-interactive)

When Phase 0.5-B grill (run by copywriter-evaluator in single-shot mode) surfaces a FATAL candidate, it cannot probe — it classifies the finding into T1 / T2 / T3 per protocols/express-mode.md §Tiered FATAL handling. Router routes based on the tier:

• T1 (AI-inferred) → ABORT Express → fall back to Q1-Q10 (interactive mode can probe).
• T2 (user-stated + benchmark-missing) → CARRY to Phase 7 with benchmark_required_before_phase_7 flag in Confirmed Assumptions. User sees the flag in Phase 0.5-C single-turn confirmation and can pre-resolve, OR Phase 7 adjudicates.
• T3 (user-stated + outright violation) → ABORT Express → fall back to Q1-Q10 (the claim itself needs discussion, not benchmark).

Tier logic exists ONLY for Express because Express lacks the interactive probe option. Q1-Q10 has no need for it.

Rationale for not unifying

Following superpowers precedent: brainstorming and subagent-driven-development are treated as two distinct skills with different return contracts, not one unified process. Interactive and non-interactive modes legitimately have different resolution strategies — unifying them would force Q1-Q10 to adopt tier labels it doesn't need, or force Express to leave questions open for a probe turn it cannot execute.

Each path is clean within its own context. The rhetoric divergence (tier vs probe) is a feature, not a bug.

Phase 0 — Brief Intake (Q1-Q10 path)

Drive the Q1-Q10 sequence in protocols/copywriting-brainstorming.md. One question per message, multiple-choice first, always attach a recommended answer. Render each question in the user's output_language (ja / en / zh-TW per §Template Rendering Rule).

Field tiers

Mirror the protocol:

Level 1 (Must — BLOCKED if missing)

• form_type — long / mid / short / ideation / audit (Q2)
• product + 1-sentence value_proposition (Q3)
• target_audience — at least 1 concrete demographic / persona (Q4)
• Form-specific must fields (Q5, branching):
• long → word-count range + Schwartz awareness level
• mid → 3+ concrete benefits + channel (Rakuten / Amazon / POP / briefing)
• short → target emotion/pain + intended channel + char-limit band
• ideation → candidate count + value-prop source
• audit → existing copy full text (not summary) + review focus

Level 2 (Should — AI recommends, user approves, defaults disclosed)

• voice_reference — 糸井 / 岩崎 / 眞木 / 谷山 / Ogilvy / 龔大中 / 許舜英 /

default (Q6). Optional voice_quadrant Q1-Q4 macro positioning.

• framework / approach — form-scoped choices (Q7): 旧/新 PASONA /

PASBECONA / BEAF / 5-approach / PREP / CREMA / Mandal-Art, etc.

Level 3 (May — off by default, opt-in)

• neta_opt_in: Yes | No (default No; Q7.5)
• neta_source_type_preference: all | sns-meme | literary | mixed (only

when opt-in Yes)

Grill (Q8)

Mandatory. Minimum one ethics-boundary question even when the brief is clean (景品表示法 / FTC / ステマ告示 risk scan). Additional probes on premises / dependencies / voice-conflict per protocol §Task 4b adaptation.

Phase 1 — Message Confirmation

After Q8 grill resolves, produce the Understanding Summary (Q9) in the exact 10-subsection structure defined in the protocol (Request / Form Type / Product / Target Audience / Form-Specific Spec / Voice Reference / Framework / Confirmed Assumptions / Resolved Ambiguities / Level 2/3 Defaults Accepted).

Decision tree — LOOSE coupling to planning-team

Before presenting the Summary to the user (Q10 confirmation gate), run:

Is the problem thesis-level?
  (unclear product positioning / audience-thesis mismatch / goal absent /
   value proposition cannot be verbalized in 1 sentence)

  YES → RECOMMEND (not enforce):
        "Consider running planning-team first for thesis clarification
         (PRODUCT-SPEC.md / JTBD / audience anchor). You may also
         proceed here — I will record the thesis as user-supplied
         and flag Level 2 defaults accordingly."
        User decides: [run planning-team first] / [proceed anyway] / [abort]
        No hard block. No auto-delegation.

  NO  → Proceed to Q10 user confirmation gate with the Summary.

Signals that trigger the RECOMMEND branch:

• Q3 value proposition cannot be written in 1 concrete sentence.
• Q4 audience is "general consumers" / "everyone" with no narrowing after

one grill round.

• User explicitly asks "what should this product even be saying?" (goal

ambiguity) rather than "how do I phrase this message?" (copy ambiguity).

Cross-plugin reference: planning-team lives in the domain-teams plugin (domain-teams:planning-team). This skill does not load it, does not pass paths to it, and does not wait on its verdict. The recommendation is a plain-English suggestion the user acts on or ignores.

Hard Gate — MUST

After Q9 Summary is drafted AND before Q10 user confirmation, dispatch the evaluator:

• Agent: copywriting-toolkit/agents/copywriter-evaluator.md
• Checklist: checklists/intake-completeness-checklist.md
• Input: the Understanding Summary artifact + form_type
• Output: verdict JSON (see checklist §Output Format)

Verdict handling:

• PASS → present Summary to user for Q10 explicit confirmation.
• PASS_WITH_NOTES (≤2 FIXABLE, no FATAL) → auto-revise once; disclose the

diff via handoff-format §Section 3 PASSWITHNOTES report; then present to user.

• NEEDS_REVISION (any FATAL, or ≥3 FIXABLE) → BLOCKED. Return to the

specific Q-task the evaluator's next_action field names. Do NOT present Summary. Do NOT proceed.

Second PASSWITHNOTES round → escalate to NEEDS_REVISION (max-2-rounds rule from handoff-format §Section 3).

User Confirmation (Q10) — Silent-proceed Prohibited

Only after gate PASS / PASSWITHNOTES AND explicit user reply of "confirm — start writing" does this skill emit the handoff envelope and return. "Adjust item" routes back to the named Q-task; "Start over" restarts at Q1.

Handoff Envelope (this skill's output)

Schema mirrors copywriting-toolkit/CLAUDE.md §Handoff Envelope:

{
  "phase": "phase-1-confirmed",
  "form": "long-form-pasona | long-form-extended | mid-form | short-form | light-action | ideation | audit",
  "brief": {
    "request": "user's paraphrased Q1 request",
    "product": "...",
    "value_proposition": "1-sentence",
    "target_audience": "demographic + Schwartz-level (long) or emotion (short)",
    "form_specific": { "word_count_range | benefits | channel | char_limit | ...": "..." },
    "voice_reference": "糸井 | 岩崎 | 眞木 | 谷山 | Ogilvy | 龔大中 | 許舜英 | default",
    "voice_quadrant": "Q1 | Q2 | Q3 | Q4 | not-declared",
    "framework": "新 PASONA | BEAF | 5-approach | CREMA | ...",
    "neta_opt_in": "Yes | No",
    "neta_source_type_preference": "all | sns-meme | literary | mixed | null",
    "level_2_3_defaults_accepted": ["voice=default (AI-recommended, user-approved)", "..."]
  },
  "message_thesis": "1-2 sentence core message derived from value prop + audience + framework",
  "planning_team_recommended": false,
  "gate_verdict": "PASS | PASS_WITH_NOTES",
  "next_stage": "copywriting-ideation | copywriting-neta-injection | copywriting-long-form-pasona | copywriting-long-form-extended | copywriting-mid-form | copywriting-short-form | copywriting-light-action | copywriting-audit-stage"
}

Downstream skills read brief + message_thesis, add their layer (ideationpool / netacandidates / draft), update next_stage, return.

Next-Step Hints

The router using-copywriting-toolkit reads next_stage and dispatches. Default routing from this skill:

• Brief is divergent / angle space unclear → copywriting-ideation (Phase 2)
• Brief is concrete + user opted neta=Yes with pre-draft bake-in →

copywriting-neta-injection (Phase 3 pre-draft)

• Brief is concrete + direct-to-draft → skip to Phase 4 workflow skill

matching form:

• long-form-pasona → copywriting-long-form-pasona
• long-form-extended (QUEST / PASTOR) → copywriting-long-form-extended
• mid-form → copywriting-mid-form
• short-form → copywriting-short-form
• light-action (PREP / CREMA) → copywriting-light-action
• form = audit → copywriting-audit-stage (bypasses Phase 2-4, runs 5-8

against existing copy)

When unsure, default next_stage to copywriting-ideation for long/short (divergence reduces backtracking) and to the Phase 4 workflow for mid/audit.

Rules (condensed — see protocol for full text)

• One question per message. Multiple choice first. Recommendation attached.
• "Don't ask what you can look up" — parse request / PRODUCT-SPEC / brand

guide before asking the user.

• Understanding Summary is a hard gate. No next-phase load without user

confirmation.

• Silent Level 2 default adoption is prohibited — disclose in Summary

§Level 2/3 Defaults Accepted.

• Voice guide > framework. If brand voice guide conflicts with Q6, voice

wins.

• Q8 grill minimum 1 question (ethics boundary) even for clean briefs.
• planning-team recommendation is LOOSE — never auto-delegate, never hard

block.

Anti-Patterns (see protocol §Anti-Patterns for examples)

• Batch questioning (Q2+Q3+Q4 in one message)
• Skipping Q8 grill
• Loading next protocol without Q10 confirmation
• Open-ended voice question ("what voice should we use?") instead of

multi-choice with representative samples

• Silent default adoption without Summary disclosure
• Re-asking what's already in the request text

Files

• protocols/copywriting-brainstorming.md — full Q1-Q10 checklist (verbatim

copy from domain-teams:copywriting-team)

• protocols/copywriting-handoff-format.md — output format + progress

reporting (verbatim copy)

• checklists/intake-completeness-checklist.md — MUST gate (verbatim copy)

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: kouko
• Source: kouko/monkey-skills
• License: MIT

Install and usage instructions live in the source repository linked above.