Copywriting Audit Stage

copywriting-audit-stage

Alternate entry point for the copywriting-toolkit pipeline. Accepts pre-existing external copy (a draft you did NOT produce — competitor copy, legacy campaigns, A/B variants, client-supplied content) and runs it through Phases 5-8 of the standard pipeline. Skips Phase 0-4 (no intake, no ideation, no neta injection, no draft) because the draft already exists.

Output: issue list by severity + suggested improvements (+ optional rewrites), with all the same gate verdicts the drafting pipeline would produce.

When to Use

• Review competitor campaigns for benchmarking
• Audit legacy copy inherited from prior agency / internal team
• Analyse A/B variants to diagnose underperformance
• Client hands you existing LP / email / banner and asks "what's wrong?"
• Pre-launch QA on copy the user produced outside this toolkit

Skip when:

• You need to generate new copy → use using-copywriting-toolkit router → full pipeline
• Audit target is technical docs / UX microcopy / PRODUCT-SPEC → out of scope (see using-copywriting-toolkit §Out of Scope)
• You need translation / localisation judgement rather than persuasion audit → research-team / design-team

Input Envelope

Accepts a compact envelope emitted by using-copywriting-toolkit audit branch:

{
  "phase": "phase-audit-entry",
  "form": "long-form-pasona | long-form-extended | mid-form | short-form | light-action | unknown",
  "brief": {
    "product": "... optional ...",
    "audience": "... optional ...",
    "voice_reference": "... optional ...",
    "review_focus": "structure | ethics | voice | all"
  },
  "external_copy": "",
  "next_stage": "copywriting-audit-stage"
}

Hard rule: external_copy must be the full original text. Audit on summaries collapses the form-adherence + voice-consistency checks.

If form is unknown, the audit protocol's Phase 1 (Type ID) determines it before downstream gates run.

Preconditions

Formal schema used by using-copywriting-toolkit router for bounce-back routing. On violation, router emits the bounce-back envelope defined in ../../CLAUDE.md §Envelope Violation. Audit-stage is an alternate entry; it does NOT require the full intake envelope.

Required envelope fields (Level 1 — BLOCKED if missing)

| Field | Type | Source | Notes | |---|---|---|---| | phase | enum | router | must equal phase-audit-entry | | external_copy | string | user | FULL TEXT of copy under review — summaries rejected | | brief.review_focus | enum | user OR default | structure / ethics / voice / all (default all) |

Optional fields (fill via defaults if absent)

| Field | Type | Notes | |---|---|---| | form | enum | if unknown, protocol Step 1 Type ID resolves it before Phase 5-8 | | brief.product | string | helpful for context; not strictly required for detection | | brief.target_audience | string | helpful for voice-consistency gate | | brief.voice_reference | string | if supplied, gates judge fidelity; if absent, audit marks voice as "declared-by-artifact" |

Upstream bounce target on violation

• external_copy missing / summary-only → bounce to using-copywriting-toolkit to re-collect full text from user
• external_copy empty / placeholder → do NOT proceed; audit has nothing to judge

Audit-stage reuses Phase 5-8 gates. If any downstream gate bounces (e.g. voice_quadrant not determinable because copy is non-prose), audit-stage halts and surfaces the block — it does NOT paper over with placeholder data.

Workflow

Uses protocols/copy-audit.md (cp'd verbatim from domain-teams/skills/copywriting-team/).

Phase 1 — Type ID            protocol Step 1   form type + framework detected
Phase 2 — Diagnose           protocol Step 2   issue list by severity
Phase 3 — Suggest            protocol Step 3   improvement recommendations
                                               (+ optional rewrites)
Phase 5 — Voice Positioning  reuse copywriting-voice-quadrant-stage
                                               assign voice_quadrant
Phase 6 — Voice Tone         reuse copywriting-voice-tone-stage
                                               tone_notes + voice-consistency
                                               SHOULD gate
Phase 7 — Ethics Gate        reuse copywriting-ethics-check-stage
                                               MUST gate (non-negotiable)
Phase 8 — Form Gate          reuse copywriting-form-check-stage
                                               MUST gate (framework adherence,
                                               length, CTA)

Phases 5-8 run on the external_copy as if it were a Phase 4 draft. They use the same evaluator (../../agents/copywriter-evaluator.md) and the same gate files — no duplicated checklists / rubrics in this skill directory.

When rewrites are produced

If Phase 3 emits rewrite variants (optional output, 1-3 variants typical), each variant undergoes an independent sub-pipeline through Phase 7 + Phase 8. This is a divergent fan-out from the main audit flow.

Variant re-gating loop — formal contract:

For each rewrite_variants[i]:

1. Construct per-variant envelope — clone the main envelope, replace draft with rewrite_variants[i].text, reset retries.revise_round_count to 0 (each variant has its own revision budget), preserve all other fields (voicequadrant from main Phase 5 result, tonenotes from Phase 6, brief unchanged).
2. Run Phase 7 ethics-check on the variant draft. Verdict → rewrite_variants[i].ethics_verdict.
3. If ethics PASS → run Phase 8 form-check. Verdict → rewrite_variants[i].form_verdict.
4. If ethics NEEDS_REVISION → skip Phase 8 for this variant; record ethics findings; variant is reported as "gate-blocked" but not auto-revised (audit does not re-draft — it reports).
5. Per-variant revise budget: revise_round_count = 2 → stop for that variant.
6. Aggregate total_retries: sum across main + all variants. If total_retries >= 4 at ANY point (main audit + variant gates combined), HALT entire audit and ask user — do not silently keep re-gating.

Execution strategy — serial by default. Parallel execution is permitted if the orchestrator supports isolated variant sub-envelopes AND guarantees retries.total_retries aggregation atomicity. Default to serial to keep the counter coherent.

Voice Consistency SHOULD gate runs cross-variant — once after all variants are evaluated, the Phase 6 voice-consistency rubric checks that the variants collectively deliver a stable voice. This is one gate per audit run, not per variant.

Rewrites follow 谷山 「なんかいいよね禁止」 — each variant must justify with 3 concrete reasons covering (a) what benefit / to whom, (b) why new vs existing copy, (c) why resonant in the given context. Description-type rationale ("it reads better" / "it's punchier") is rejected; the reason must name a concrete audience / benefit / rhetorical mechanism. This rule is applied locally here; audit does not delegate to copywriting-ideation.

Output

Audit report envelope:

{
  "phase": "audit-complete",
  "form_detected": "long-form-pasona | ...",
  "framework_detected": "pasona | bf | quest | crema | ...",
  "issues": [
    {
      "severity": "critical | high | medium | low",
      "category": "structure | ethics | voice | length | cta",
      "location": "",
      "finding": "...",
      "suggestion": "..."
    }
  ],
  "rewrite_variants": [
    {
      "variant_id": "A",
      "text": "...",
      "three_reasons": ["...", "...", "..."],
      "ethics_verdict": "PASS | PASS_WITH_NOTES | NEEDS_REVISION",
      "form_verdict": "PASS | PASS_WITH_NOTES | NEEDS_REVISION"
    }
  ],
  "voice_quadrant": { "primary": "Q1 | Q2 | Q3 | Q4", "edge": "...", "rationale": "...", "schwartz_alignment": "..." },
  "tone_notes": "...",
  "ethics_verdict_on_original": "PASS | PASS_WITH_NOTES | NEEDS_REVISION",
  "form_verdict_on_original": "PASS | PASS_WITH_NOTES | NEEDS_REVISION",
  "next_stage": null
}

Loop-back

If the user asks for revised rewrites (not just findings):

• Route rewrites to copywriting-voice-tone-stage → copywriting-ethics-check-stage → copywriting-form-check-stage (reuse the main pipeline's loop-back cap: 2 revise rounds per variant, then human escalation).
• Do NOT re-enter Phase 4 drafter skills from audit — if the user wants a full rewrite from scratch, exit audit and use using-copywriting-toolkit to enter the standard pipeline.

Hard Rules

• Full text, not summary — audit on summaries is invalid.
• Original copy is never modified — audit produces reports + optional variants; the original survives unchanged in the envelope.
• Ethics gate verdict on original is reported even if NEEDS_REVISION — a failing audit is a valid audit; the user needs to know the legal / ethical exposure.
• Framework detection before gates — form-check-stage cannot run without knowing the framework. If the copy resists classification, the protocol's Phase 1 returns "framework_detected: unknown" and form-check-stage runs in exploratory mode per its own §Unknown-form handling.
• Do NOT duplicate gate files here — this skill reuses copywriting-ethics-check-stage/ + copywriting-form-check-stage/ + copywriting-voice-tone-stage/ gates via the orchestrator. Drift-sync risk avoided.

Next Stage

• On completion: return audit report to user. No downstream stage.
• On user request for revised variants: enter the Phase 6 → 7 → 8 sub-loop described above.

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.