greenwashing-self-audit
GitHub用于在发布前审查营销内容中的环保声明，识别漂绿风险。通过提取声明、评估证据充分性、标记模糊术语及遗漏项，为每项声明提供修改或删减建议，帮助规避监管或舆论危机。
Trigger Scenarios
Install
npx skills add mohitagw15856/pm-claude-skills --skill greenwashing-self-audit -g -y
SKILL.md
Frontmatter
{
"name": "greenwashing-self-audit",
"description": "Audit your own marketing and report claims for greenwashing risk before a regulator, journalist, or competitor does. Use when asked to review sustainability claims, check marketing copy for greenwash, audit environmental claims on a website or report, or pressure-test green messaging. Produces a claim inventory with substantiation status per claim, vague-term and omission flags, and a fix-or-drop recommendation for every claim."
}
Greenwashing Self-Audit Skill
The cheapest time to find an unsupportable green claim is before publication; the most expensive is in a regulator's letter. This skill inventories every environmental claim in a piece of content, grades its substantiation, flags vague terms and material omissions, and ends with a concrete fix-or-drop call per claim. Audit the material as an adversary would read it, not as the author meant it.
What This Skill Produces
- A numbered claim inventory extracted from the material (explicit claims and implied ones)
- A substantiation status per claim: Evidenced / Partially evidenced / Unsupported
- Vague-term flags with the precise question each term fails to answer
- An omission review — what a full picture would include that the material leaves out
- A fix-or-drop recommendation per claim, with rewritten wording where a fix exists
Required Inputs
Ask for these if not provided; if only the copy is supplied, run the audit anyway and mark every substantiation status as "pending evidence" rather than refusing:
- The material — the marketing copy, report section, packaging text, or web page content
- Available evidence — data, certifications, methodologies, or studies behind the claims
- Claim scope — does each claim cover a product, a product line, or the whole company?
- Audience and jurisdiction — consumer-facing vs B2B, and where it will be published (affects which rules bite)
Audit Framework
1. Build the claim inventory. Extract every environmental claim: explicit statements, numbers, badges/imagery (green leaves, "planet-friendly" visuals count), and implied claims (a wind-turbine photo on a fossil product page is a claim). Number each.
2. Grade substantiation per claim:
| Status | Test |
|---|---|
| Evidenced | Specific, current, methodology-backed evidence exists and covers the claim's full scope |
| Partially evidenced | Evidence exists but is narrower, older, or weaker than the claim as worded |
| Unsupported | No evidence provided, or evidence contradicts the claim's scope or tense |
Grade the claim as a skeptical reader would parse it, not as the author intended it.
3. Flag vague terms. "Eco-friendly", "green", "sustainable", "natural", "climate positive", "carbon neutral" (without standard + registry), "recyclable" (without where/how), "up to X%" (without typical case). For each: what precise question does the term dodge?
4. Run the omission review. Would the claim survive full context? Check for: cherry-picked scope (one green product fronting a brown portfolio), lifecycle stages ignored (use-phase highlighted, manufacturing hidden), trade-offs unmentioned (recyclable but carbon-heavy), progress framed against an unstated or convenient baseline, and offsets doing silent work in a "neutral" claim.
5. Decide fix-or-drop per claim. A claim is fixable if narrowing scope, adding basis, or quantifying makes it true and evidenced ("100% recycled packaging on product X, certified by [scheme]"). Otherwise recommend dropping it. Provide the rewritten wording for every "fix".
Output Format
Greenwashing self-audit: [material name]
1. Summary risk read — overall exposure (low / moderate / high), the riskiest 2–3 claims, and the dominant failure mode (vagueness, scope inflation, offset reliance, omission).
2. Claim inventory and grades — table: # | claim (verbatim) | type (explicit / implied / visual) | scope | substantiation status | evidence cited.
3. Vague-term flags — term | where it appears | the unanswered question | suggested precise replacement.
4. Omission review — what a complete picture adds, and which claims it would change.
5. Fix-or-drop register — per claim: Fix (with rewritten wording and the evidence to attach) or Drop (with the reason a fix isn't available).
6. Evidence to-collect list — what proof, if obtained, would upgrade "partially evidenced" claims.
Include this line in the artifact: "Verify final claims against the applicable advertising and disclosure rules in your jurisdictions (e.g. green claims guidance, unfair commercial practices rules) with your legal and compliance team before publication."
Quality Checks
- Implied and visual claims are inventoried, not just literal sentences
- Every claim has a substantiation status and the evidence (or its absence) is named
- Every vague term is flagged with the specific question it fails to answer
- The omission review considers scope, lifecycle, trade-offs, baseline, and offsets
- Every claim ends in an explicit Fix (with new wording) or Drop — no "monitor" limbo
- The audit reads adversarially — graded as a regulator or journalist would parse it
Anti-Patterns
- Do not claim carbon neutrality without naming the standard, the offset registry, and the gross-emissions figure it sits on
- Do not grade a claim by author intent — grade the words as published
- Do not let a true-but-narrow claim imply company-wide performance — scope must match wording
- Do not soften an "Unsupported" grade to "Partially" out of politeness — the reader outside won't
- Do not fix a claim by making it vaguer — fixes add basis and precision, or the claim drops
- Do not treat this audit as legal clearance — jurisdictional rules and counsel govern
Based On
Green-claims substantiation practice (claim–evidence matching, scope discipline, omission and lifecycle review) as used in advertising-standards and consumer-protection contexts.
Version History
- 54fad50 Current 2026-07-19 13:29
