{"slug":"check","title":"check","summary":"Run the scored pre-publish gate via eval-runner.py: claims, brand voice, compliance, AI tells. \"check this before we publish\"","platform":"Claude","tags":[],"authorName":"LLM Mart","authorSlug":"llm-mart","score":0,"source":"github","price":null,"verified":false,"createdAt":"2026-10-27T18:22:19.733582Z","repo":{"url":"https://github.com/indranilbanerjee/digital-marketing-pro","stars":968,"forks":160,"license":"MIT","updatedAt":"2026-11-07T17:49:28Z"},"bodyHtml":"<hr>\n<h2>name: check\ndescription: \"Run the scored pre-publish gate via eval-runner.py: claims, brand voice, compliance, AI tells. \"check this before we publish\"\"\nuser-invocable: true\nargument-hint: \"</h2>\n<h1>/digital-marketing-pro:check — Unified Pre-Publish Quality Gate</h1>\n<blockquote>\n<p><strong>Script location.</strong> If your host does not set <code>${CLAUDE_PLUGIN_ROOT}</code>, the scripts are in this plugin's <code>scripts/</code> folder, next to <code>skills/</code>. Run them with Python 3.10 or newer: <code>python</code>, or <code>python3</code> where <code>python</code> is not found (macOS, many Linux systems).</p>\n</blockquote>\n<p>This skill is the canonical pre-publish gate for marketing content. It wraps the evaluation suite (<code>scripts/eval-runner.py</code>) and produces a single pass/fail decision with actionable issues.</p>\n<h2>Context efficiency</h2>\n<p>Heavy skill. <strong>Grep before Read</strong> any referenced file, then <code>Read</code> only matched ranges with <code>offset</code> + <code>limit</code>. List the brand's workspace at <code>~/.claude-marketing/brands/{slug}/</code> (or <code>$CLAUDE_PLUGIN_DATA/digital-marketing-pro/brands/{slug}/</code> when that env var is set) before opening files. On re-invocation mid-session, skip files already in context.</p>\n<p>Use this skill <strong>before publishing any marketing content</strong> — blog posts, ad copy, emails, social posts, landing pages, press releases, or any branded copy.</p>\n<h2>Why this skill exists</h2>\n<p>An earlier version shipped a global PreToolUse hook that auto-ran a hallucination + brand-compliance check on every Write/Edit operation in every project. That hook was removed because it fired globally across all plugins and projects (Slack writes, GitHub PRs, code edits — all of it), causing friction in non-marketing work.</p>\n<p><code>/digital-marketing-pro:check</code> replaces that automatic gate with an <strong>explicit user-invoked gate</strong>. The work is the same; the trigger is intentional.</p>\n<h2>What the check evaluates</h2>\n<p>The check delegates to <code>scripts/eval-runner.py</code> (the master eval orchestrator) which calls four sibling scripts:</p>\n<table>\n<thead>\n<tr>\n<th>Dimension</th>\n<th>Script</th>\n<th>What it checks</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td><strong>Hallucination</strong></td>\n<td><code>hallucination-detector.py</code></td>\n<td>Unattributed statistics, placeholder URLs (example.com / your-site.com), unsupported superlatives (\"best\", \"#1\", \"leading\"), fabricated citations</td>\n</tr>\n<tr>\n<td><strong>Claims</strong></td>\n<td><code>claim-verifier.py</code> (when <code>--evidence</code> provided)</td>\n<td>Cross-checks specific claims against a user-provided evidence file</td>\n</tr>\n<tr>\n<td><strong>Brand voice</strong></td>\n<td><code>brand-voice-scorer.py</code> (when <code>--brand</code> provided)</td>\n<td>Scores content against the active brand's voice profile (formality, energy, humor, authority, prefer/avoid words)</td>\n</tr>\n<tr>\n<td><strong>Structure</strong></td>\n<td><code>output-validator.py</code> (when <code>--schema</code> provided)</td>\n<td>Validates content matches expected schema (blog_post, email, ad_copy, social_post, landing_page, press_release, content_brief, campaign_plan)</td>\n</tr>\n<tr>\n<td><strong>C2PA provenance</strong> (compliance)</td>\n<td><code>embed-c2pa.py</code> (presence check)</td>\n<td>When the brand's <code>target_markets</code> include an EU/EEA jurisdiction AND an accompanying asset is AI-generated: verifies a C2PA provenance manifest is present and valid. Missing or invalid manifest → <strong>CRITICAL / BLOCKED</strong> (EU AI Act Article 50, applies from 2 Aug 2026)</td>\n</tr>\n</tbody>\n</table>\n<p>Plus content quality and readability scoring (always run).</p>\n<h2>Subcommands and modes</h2>\n<h3>Default (run-quick)</h3>\n<pre><code>/digital-marketing-pro:check &lt;file-path-or-content&gt;\n</code></pre>\n<p>Runs the <strong>quick eval</strong>: hallucination detection + content quality + readability. Fast (~2 seconds), zero external dependencies. Use this for routine checks.</p>\n<h3>Full eval (run-full)</h3>\n<pre><code>/digital-marketing-pro:check &lt;file-path-or-content&gt; --full\n</code></pre>\n<p>Runs all 6 dimensions: hallucination + claims (if evidence provided) + brand voice (if brand provided) + structure (if schema provided) + content quality + readability. Use before publishing anything client-facing or external.</p>\n<h3>Compliance-focused (run-compliance)</h3>\n<pre><code>/digital-marketing-pro:check &lt;file-path-or-content&gt; --compliance --brand &lt;slug&gt; [--evidence &lt;path&gt;] [--schema &lt;name&gt;]\n</code></pre>\n<p>Runs hallucination + claims + brand voice + structure. Best for regulated industries (healthcare, financial services, alcohol, cannabis, gambling) where claim substantiation and brand-voice fidelity matter most.</p>\n<h3>With evidence file</h3>\n<pre><code>/digital-marketing-pro:check &lt;file-path&gt; --evidence &lt;evidence-file.json&gt;\n</code></pre>\n<p>When the content makes specific claims you want to substantiate, provide a JSON evidence file:</p>\n<pre><code>{\n  \"evidence\": [\n    {\n      \"claim\": \"50% increase in conversions\",\n      \"source\": \"GA4 Q4 report\",\n      \"date\": \"2025-12-31\",\n      \"verified\": true\n    },\n    {\n      \"claim\": \"Trusted by Fortune 500 companies\",\n      \"source\": \"Customer roster (internal)\",\n      \"date\": \"2026-04-01\",\n      \"verified\": true\n    }\n  ]\n}\n</code></pre>\n<p>The check will extract every claim from the content and flag any that don't match an evidence entry.</p>\n<h3>With schema validation</h3>\n<pre><code>/digital-marketing-pro:check &lt;file-path&gt; --schema blog_post\n</code></pre>\n<p>Validates the content matches the structural requirements of the named schema. Available schemas: <code>blog_post</code>, <code>email</code>, <code>ad_copy</code>, <code>social_post</code>, <code>landing_page</code>, <code>press_release</code>, <code>content_brief</code>, <code>campaign_plan</code>. Use <code>--schema list</code> to see all schemas with their requirements.</p>\n<h3>With brand voice check</h3>\n<pre><code>/digital-marketing-pro:check &lt;file-path&gt; --brand acme\n</code></pre>\n<p>Scores the content against the brand voice profile at <code>~/.claude-marketing/brands/acme/profile.json</code>. Reports per-dimension breakdown (formality, energy, humor, authority) plus deviation from prefer/avoid word lists.</p>\n<h2>Output format</h2>\n<p>The check returns a unified report:</p>\n<pre><code>DM CHECK REPORT — &lt;file or content snippet&gt;\n=============================================\n\nComposite Score: 73.4 / 100  (Grade: B-)\nAuto-Reject: NO\n\nDimensions:\n  Hallucination ............ 96/100  PASS  (weight 0.40)\n  Content Quality .......... 78/100  PASS  (weight 0.35)\n  Readability .............. 65/100  PASS  (weight 0.25)\n\nIssues Found:\n  CRITICAL: None\n  WARNING (2):\n    - Line 14: Unattributed statistic \"76% of buyers prefer...\"\n      Suggestion: cite source or rephrase as observation\n    - Line 22: Superlative \"best in class\" without substantiation\n      Suggestion: replace with measurable claim or proof point\n\nDecision: PASS — safe to publish but address WARNINGs first\n</code></pre>\n<p>If any CRITICAL issue is found, decision = <strong>BLOCKED</strong> and the user is asked to fix before publishing.</p>\n<h3>Decision rules</h3>\n<ul>\n<li><strong>PASS</strong> — no CRITICAL issues and the composite score is above the auto-reject threshold (<code>auto_reject_threshold</code>, default 40)</li>\n<li><strong>WARN</strong> — no CRITICAL issues but at least one WARNING; the user should address it before publishing</li>\n<li><strong>BLOCKED</strong> — at least one CRITICAL issue (e.g. placeholder URL, fabricated statistic in a headline, missing required disclaimer for a regulated industry, missing C2PA provenance manifest on an AI-generated asset in an EU-targeted campaign); the content cannot publish until fixed</li>\n</ul>\n<h2>AI-tell scans (advisory section, never scored)</h2>\n<p>Alongside the eval-runner scorers, run both tell scans and report them as a single ADVISORY section of the check output:</p>\n<pre><code>python \"${CLAUDE_PLUGIN_ROOT}/scripts/ai-tell-scan.py\"          --file &lt;input&gt;   # Tier 1: surface\npython \"${CLAUDE_PLUGIN_ROOT}/scripts/structural-tell-scan.py\"  --file &lt;input&gt;   # Tier 2: structure\n</code></pre>\n<ul>\n<li><strong>Tier 1 (surface)</strong> — LLM-favored vocabulary, significance markers, soft-adverb clusters, connective and participial openers, em-dash density, ungrounded one-liners. Report the overall LOW/MODERATE/HIGH rating and the flagged sentences with their suggested fix. Significance markers are reported with <code>\"fix\": \"Delete this sentence; do not reword it.\"</code> — pass that through verbatim, because rewording is the wrong remedy.</li>\n<li><strong>Tier 2 (structure)</strong> — the overall OK/NOTE/ATTENTION band plus each NOTE/ATTENTION finding with its spans (moralizing, section symmetry, parallel headings, specificity, stance, paragraph evenness, entity development). For <code>entity_development</code>, always carry through that the fix is to develop an existing specific, never to delete specifics.</li>\n</ul>\n<p><strong>This whole section NEVER affects the PASS/WARN/BLOCKED decision.</strong> Both scripts keep their thresholds inside themselves, deliberately outside the eval config, because these are editorial judgment calls for a human editor, not publish gates — and because a detector proxy has a real false-positive rate on genuinely human writing. (The one place a tell scan does gate is the content-engine's <code>humanize_passed</code>, and only on the two tells precise enough to gate on: <code>significance_marker</code> and <code>soft_adverb_cluster</code>. <code>llm_favored_word</code> was dropped from that set on 2026-08-15 after it was measured firing <strong>only</strong> on prose published before ChatGPT existed and never on model prose. That gate is a density floor — measured, it fails no published human writing and catches no unedited model prose — so never report a pass as evidence that a piece reads human.) Both scans measure visible text only; neither can see, and neither has any relationship to, any statistical watermark.</p>\n<h2>EU AI Act Article 50 — C2PA provenance gate</h2>\n<p>The check gains a compliance dimension for AI-generated assets in EU-targeted campaigns. It fires when <strong>both</strong> conditions hold:</p>\n<ol>\n<li>The active (or <code>--brand</code>) profile's <code>target_markets</code> include any EU/EEA jurisdiction, <strong>and</strong></li>\n<li>An accompanying asset is declared AI-generated — either the file metadata says so, or the <code>--evidence</code> JSON declares <code>ai_generated: true</code> for it.</li>\n</ol>\n<p>When both hold, the gate checks the asset read-only with <code>python \"${CLAUDE_PLUGIN_ROOT}/scripts/embed-c2pa.py\" --verify &lt;asset&gt;</code>: exit 0 = manifest present and valid (a self-signed dev signer is reported as <code>untrusted_signer</code>, not failed), 6 = no manifest, 7 = manifest fails validation, 2 = c2pa-python not installed (report the dimension as SKIPPED with the printed install command; never PASS it). A <strong>missing or invalid C2PA provenance manifest (exit 6 or 7) is a CRITICAL issue → decision = BLOCKED.</strong> Article 50 applies from <strong>2 Aug 2026</strong> (penalty up to EUR 15M or 3% of global turnover). To embed a compliant manifest, run <code>/digital-marketing-pro:c2pa-metadata</code>.</p>\n<p>If <code>embed-c2pa.py</code> is not present in the script inventory or the asset cannot be resolved, surface the dimension as SKIPPED with a warning (never silently PASS an EU AI-asset check).</p>\n<h2>How the skill operates</h2>\n<p>The skill follows this flow:</p>\n<ol>\n<li><strong>Resolve the input.</strong> If the user passed a file path, read it. If they passed inline content, use it.</li>\n<li><strong>Resolve options.</strong> If <code>--brand</code> not specified, attempt to load from active brand at <code>~/.claude-marketing/brands/_active-brand.json</code>. If <code>--schema</code> not specified, infer from content type if obvious (blog markdown → <code>blog_post</code>, etc.) or skip structure check.</li>\n<li><strong>Build the eval-runner command.</strong> Choose action: <code>run-quick</code> (default), <code>run-full</code> (with <code>--full</code>), <code>run-compliance</code> (with <code>--compliance</code>).</li>\n<li><strong>Execute via Bash.</strong>\n<pre><code>python \"${CLAUDE_PLUGIN_ROOT}/scripts/eval-runner.py\" --action run-quick --file &lt;input&gt; [--brand &lt;slug&gt;] [--evidence &lt;path&gt;] [--schema &lt;name&gt;]\n</code></pre>\n</li>\n<li><strong>Parse the JSON output.</strong> Extract composite score, grade, dimension scores, alerts, auto-reject decision.</li>\n<li><strong>Format for the user.</strong> Present the human-readable report shown above. Lead with the decision (PASS / WARN / BLOCKED).</li>\n<li><strong>If BLOCKED, refuse to recommend publishing.</strong> Always require the user to address CRITICAL issues before they proceed.</li>\n</ol>\n<h2>Scripts called</h2>\n<ul>\n<li><code>scripts/eval-runner.py</code> — master orchestrator</li>\n<li><code>scripts/hallucination-detector.py</code> — invoked by eval-runner</li>\n<li><code>scripts/claim-verifier.py</code> — invoked by eval-runner if <code>--evidence</code> provided</li>\n<li><code>scripts/brand-voice-scorer.py</code> — invoked by eval-runner if <code>--brand</code> provided</li>\n<li><code>scripts/output-validator.py</code> — invoked by eval-runner if <code>--schema</code> provided</li>\n<li><code>scripts/content-scorer.py</code> — invoked by eval-runner</li>\n<li><code>scripts/readability-analyzer.py</code> — invoked by eval-runner</li>\n<li><code>scripts/embed-c2pa.py</code> — presence/verify check for the EU AI Act Article 50 C2PA gate (only when an EU-targeted brand has an AI-generated asset)</li>\n</ul>\n<p>All scripts use stdlib only (except brand-voice-scorer which optionally uses nltk). No external API calls, no internet required.</p>\n<h2>Examples</h2>\n<h3>Example 1: Quick check on a draft</h3>\n<pre><code>User: /digital-marketing-pro:check drafts/q2-launch-blog.md\n\nSkill:\n1. Read drafts/q2-launch-blog.md\n2. Run python \"${CLAUDE_PLUGIN_ROOT}/scripts/eval-runner.py\" --action run-quick --file drafts/q2-launch-blog.md\n3. Parse JSON output:\n   composite_score: 81.2, grade: B+, auto_rejected: false\n   hallucination: 92/100 pass, content_quality: 76/100 pass, readability: 84/100 pass\n   alerts: 1 warning (\"unattributed stat in line 14\")\n4. Format report:\n\nDM CHECK REPORT — drafts/q2-launch-blog.md\n============================================\nComposite Score: 81.2 / 100  (Grade: B+)\nDecision: PASS\n\nDimensions:\n  Hallucination ......... 92/100  pass\n  Content Quality ....... 76/100  pass\n  Readability ........... 84/100  pass\n\nIssues Found:\n  WARNING (1):\n    - Line 14: Unattributed statistic \"76% of marketers say...\"\n      Suggestion: cite source or rephrase as observation\n\nDecision: PASS — safe to publish; recommend addressing the WARNING first.\n</code></pre>\n<h3>Example 2: Full eval with brand + evidence + schema</h3>\n<pre><code>User: /digital-marketing-pro:check drafts/healthcare-ad.md --full --brand healthfirst --evidence facts/q2-claims.json --schema ad_copy\n\nSkill:\n1. Read drafts/healthcare-ad.md\n2. Run python \"${CLAUDE_PLUGIN_ROOT}/scripts/eval-runner.py\" --action run-full --file drafts/healthcare-ad.md --brand healthfirst --evidence facts/q2-claims.json --schema ad_copy\n3. Parse JSON output. Composite: 58.4, grade: D+, auto_rejected: true\n4. Format report with CRITICAL issues highlighted\n5. Decision: BLOCKED. Two unattributed health claims need substantiation before this can publish.\n</code></pre>\n<h3>Example 3: Compliance check on regulated content</h3>\n<pre><code>User: /digital-marketing-pro:check drafts/financial-services-landing.md --compliance --brand finadvisor --evidence facts/finra-disclosures.json\n\nSkill:\n1. Read content\n2. Run python \"${CLAUDE_PLUGIN_ROOT}/scripts/eval-runner.py\" --action run-compliance --file drafts/financial-services-landing.md --brand finadvisor --evidence facts/finra-disclosures.json\n3. Output prioritises hallucination + claim verification + brand voice + structure\n4. Returns decision with FINRA-relevant issues highlighted\n</code></pre>\n<h3>Example 4: Quick check on inline content</h3>\n<pre><code>User: /digital-marketing-pro:check \"Our amazing product boosts conversion by 347% — visit example.com today!\"\n\nSkill:\n1. Detect inline content (not a file path)\n2. Write content to a temp file\n3. Run quick eval\n4. Report:\n   CRITICAL: 2\n     - Placeholder URL \"example.com\" — replace with real URL before publishing\n     - Unattributed statistic \"347%\" — fabricated stat or missing citation\n   Decision: BLOCKED\n</code></pre>\n<h2>When to use which mode</h2>\n<table>\n<thead>\n<tr>\n<th>Scenario</th>\n<th>Recommended mode</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>Routine content check during drafting</td>\n<td><code>/digital-marketing-pro:check &lt;file&gt;</code> (quick)</td>\n</tr>\n<tr>\n<td>Before publishing any external content</td>\n<td><code>/digital-marketing-pro:check &lt;file&gt; --full --brand &lt;slug&gt;</code></td>\n</tr>\n<tr>\n<td>Regulated industry content (healthcare / financial / alcohol / cannabis / gambling)</td>\n<td><code>/digital-marketing-pro:check &lt;file&gt; --compliance --brand &lt;slug&gt; --evidence &lt;facts&gt;</code></td>\n</tr>\n<tr>\n<td>Client-facing deliverable (Growth Plan, Yearly Planner, monthly report)</td>\n<td><code>/digital-marketing-pro:check &lt;file&gt; --full --brand &lt;slug&gt;</code></td>\n</tr>\n<tr>\n<td>Ad copy specifically</td>\n<td><code>/digital-marketing-pro:check &lt;file&gt; --schema ad_copy --brand &lt;slug&gt;</code></td>\n</tr>\n<tr>\n<td>Email specifically</td>\n<td><code>/digital-marketing-pro:check &lt;file&gt; --schema email --brand &lt;slug&gt;</code></td>\n</tr>\n<tr>\n<td>Blog post specifically</td>\n<td><code>/digital-marketing-pro:check &lt;file&gt; --schema blog_post --brand &lt;slug&gt;</code></td>\n</tr>\n</tbody>\n</table>\n<h2>Behaviour rules</h2>\n<ol>\n<li><strong>Never report PASS if there are CRITICAL issues.</strong> Always BLOCKED.</li>\n<li><strong>Always report the composite score and grade.</strong> Even if PASS, surface room for improvement.</li>\n<li><strong>Always include actionable suggestions.</strong> Each issue must be paired with a fix recommendation.</li>\n<li><strong>Resolve the active brand if not specified.</strong> Check <code>~/.claude-marketing/brands/_active-brand.json</code>. If no active brand, run without <code>--brand</code> (skip brand voice dimension).</li>\n<li><strong>Never modify the content.</strong> This skill only reports — the user (or the agent that produced the content) makes the fix.</li>\n<li><strong>Surface skipped dimensions explicitly.</strong> If the user did not provide <code>--evidence</code> or <code>--schema</code>, note that the corresponding dimensions were skipped.</li>\n</ol>\n<h2>Related skills + commands</h2>\n<ul>\n<li><code>/digital-marketing-pro:engagement growth-plan</code> — produces Part 8 deliverable; should be checked with <code>/digital-marketing-pro:check --full --schema content_brief</code> before client delivery</li>\n<li><code>/digital-marketing-pro:content-engine</code> — produces marketing content; recommended workflow is <code>/digital-marketing-pro:content-engine</code> → review → <code>/digital-marketing-pro:check</code> → publish</li>\n<li><code>/digital-marketing-pro:eval-content</code> — legacy alias that routes to this skill</li>\n</ul>\n<h2>Related references</h2>\n<ul>\n<li><code>scripts/eval-runner.py</code> — the master orchestrator this skill wraps</li>\n<li><code>skills/context-engine/eval-framework-guide.md</code> — full eval framework documentation</li>\n<li><code>skills/context-engine/eval-rubrics.md</code> — per-dimension scoring rubrics</li>\n<li><code>docs/architecture.md</code> Section 16 (Evaluation Layer) — eval framework architecture</li>\n</ul>\n","files":[{"path":"SKILL.md","sizeBytes":22328,"isText":true}],"reviewScore":null,"reviewSummary":null,"trust":{"provenance":"trusted-source-unreviewed","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow.","bodySource":null},"bodyLocked":false,"purchaseUrl":null,"sourceUrl":null,"report":{"provenance":"trusted-source-unreviewed","screen":{"ran":true,"outcome":"clean","suspicious":0,"notes":0,"hiddenCharacters":false},"virusScan":{"engine":"clamav","status":"clean","scannedAt":"2026-11-17T00:45:11.299493Z","sha256":"0D4000CACC5EC75DB3A6B0C3ECBD15ECBD7F75C414C7F7197F25FF01AD24E2CF","sizeBytes":8420},"review":null,"source":{"repositoryUrl":"https://github.com/indranilbanerjee/digital-marketing-pro","path":"skills/check","license":"MIT","commit":"be4142b25bba6d424962a7788432dd813e6b768c","subtreeSha":"FACE7FD63885D603CE31AB5DC120D17B06FBD24EC0A53FD50E8D4AB190D3673A","lastSyncedAt":"2026-11-17T00:40:28.514301Z"},"reviewedAt":"2026-11-17T00:54:54.70961Z","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow."},"install":[{"target":"skills-cli","command":"npx skills add https://github.com/indranilbanerjee/digital-marketing-pro/tree/main/skills/check"},{"target":"claude-code","command":"claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install indranilbanerjee-digital-marketing-pro@llmmart"},{"target":"git","command":"git clone https://github.com/indranilbanerjee/digital-marketing-pro.git"}]}