Social media and marketing workflows

Input:

Draft: @AcmeLabs We shipped the new report today https://example.com/very-long-article #launch 🚀

Output:

Revised draft: We shipped the new report today for Acme Labs. Read the full article: https://example.com/very-long-article #launch 🚀

Check:

The rewrite removes the leading mention from the start, keeps the link and emoji intact for composer validation, and preserves the message without claiming any special weighted character rule.

Input:

Alt text draft: photo of a happy team at the office with a chart behind them

Output:

Revised alt text: team members standing in an office beside a bar chart on a wall

Check:

The revision drops the subjective word happy and keeps only observable details. It is framed as alt text only, without inventing a platform-specific limit or hidden metadata.

Input:

Caption: New event details below. Alt text: #OpenDataDay poster at the library. Hashtags: open data day, civic tech

Output:

Revised caption: New event details below. Hashtags: #OpenDataDay #CivicTech. Cleaned alt text: poster at a library

Check:

The hashtags are moved out of alt text, the multiword tags are normalized in CamelCase, and the alt text stays descriptive without adding unsupported details such as a date or specific venue name.

Input:

Chart type: line chart. Takeaway: sales rose after July. Extremes: 120 in January, 240 in September. Labels: months, sales in USD.

Output:

Alt text: line chart showing sales rising after July, from 120 in January to 240 in September, with months on the x-axis and sales in USD on the y-axis.

Check:

The description covers the chart type, overall trend, and the supplied values without listing a full dataset. It does not add any unsupported map or location details.

Input:

Synthetic source excerpt: "Battery recycling begins with collection at designated drop-off sites. Facilities then sort batteries by chemistry because different battery types require different recovery processes. Recovery facilities extract reusable materials such as lithium, nickel, and cobalt." Need 3 X posts.

Output:

Post 1: Battery recycling starts with collection at designated drop-off sites.

Post 2: Facilities sort batteries by chemistry because different battery types require different recovery processes.

Post 3: Recovery facilities can extract reusable materials such as lithium, nickel, and cobalt.

Check:

Each post is a concrete derivative of the synthetic source excerpt, contains only supplied facts, and is under 280 characters.

Input:

Invented illustrative data: event_datetime=2026-11-10 21:00, event_iana_zone=America/New_York, operator_iana_zone=Europe/Berlin

Output:

The same instant is 2026-11-11 02:00 UTC and 2026-11-11 03:00 in Europe/Berlin, while it remains 2026-11-10 21:00 in America/New_York.

Check:

This corrects the timezone math: 21:00 in New York on 2026-11-10 is 02:00 UTC the next day and 03:00 in Berlin. The dates and offsets are consistent.

Input:

Invented illustrative data: weekday=Thursday, civil_time=20:00, anchor_iana_zone=America/New_York, start_date=2026-10-29

Output:

The first two occurrences are 2026-10-29 20:00 EDT and 2026-11-05 20:00 EST, with different UTC offsets because daylight saving ends between those dates.

Check:

Both entries keep the same local clock time, and the abbreviation changes from EDT to EST after the DST transition. No unsupported scheduling aliasing or automation promise is implied.

Input:

Synthetic data: publisher_local_datetime=2026-10-02 19:00, publisher_iana_zone=Asia/Tokyo, viewer_iana_zones=[America/New_York, America/Los_Angeles]

Output:

2026-10-02 19:00 in Asia/Tokyo is 2026-10-02 10:00 UTC. The same instant is 2026-10-02 06:00 in America/New_York and 2026-10-02 03:00 in America/Los_Angeles.

Check:

Tokyo is UTC+9 throughout the year. On 2026-10-02, New York is UTC-4 and Los Angeles is UTC-7, so Tokyo is 13 hours ahead of New York and 16 hours ahead of Los Angeles.

Input:

Synthetic example: Post text is "Launch note." Proposed publication time is 2026-10-01 16:00 UTC. Publication requires Mina's manual approval.

Output:

MANUAL APPROVAL CHECKLIST
[ ] Mina reviewed the exact text: "Launch note."
[ ] Mina confirmed the intended account.
[ ] Mina confirmed the publication time: 2026-10-01 16:00 UTC.
[ ] Mina approved publication.

Do not schedule or publish unless every item is manually checked.

Check:

The output provides a manual approval gate and does not claim that an approver, deadline, approval state, or schedule was stored.

Input:

Synthetic CSV row: body=Hello, publish_at=2026-10-06T09:00:00-04:00, network=x

Output:

Use publish_at=2026-10-06T09:00:00-04:00. Its UTC equivalent is 2026-10-06T13:00:00Z.

Check:

The timestamp includes an explicit UTC-4 offset. Adding four hours to 09:00 gives 13:00 UTC on the same date. The example does not claim that an unspecified importer accepted the row.

Input:

Invented illustrative data: platform=x, region_or_timezone=US Eastern, date_range=2026-10-01 to 2026-10-31

Output:

Illustrative guidance only: use US Eastern as a baseline if your audience is nationwide, then compare it with account analytics when available.

Check:

The wording is clearly nonexclusive and nonpersonalized, and it does not pretend to return ranked candidate windows or hidden analytics results.

Input:

{"robots_txt":"User-agent: *\nDisallow: /private/\nSitemap: https://example.com/sitemap.xml","url":"https://example.com/private/report","goal":"exclude","html_head":"<head><title>Quarterly Report</title><meta name=\"robots\" content=\"index,follow\"></head>","x_robots_tag":null}

Output:

{"urlAllowed":false,"matchedRule":"Disallow: /private/","explanation":"Google cannot crawl the blocked URL path, so it may show limited information or report that a description is unavailable because of robots.txt.","recommendedFix":"First determine whether the page is public or private. If it is public but should be excluded from search, allow crawling and apply noindex on the page or via an HTTP header. If it is private, require authentication rather than relying on robots.txt or noindex.","notes":["robots.txt alone is not a reliable removal or access-control method"]}

Check:

The path is blocked by robots.txt, so the page is not crawlable. The output correctly distinguishes a public page that should use noindex from private content that should require authentication, without treating robots.txt as reliable removal or access control.

Input:

{"title":"Home | Example Site","h1":"Blue Widget Installation Guide","og_title":"Home | Example Site","site_name":"Example Site","language":"en","approved_terms":["Blue Widget"]}

Output:

{"issues":["Title is generic","Title repeats site boilerplate","Title does not match the page topic"],"proposedTitle":"Blue Widget Installation Guide | Example Site","supportingSignals":["H1 mentions Blue Widget Installation Guide","Approved term Blue Widget is present"],"manualChecks":["Confirm the title is unique on the site","Verify the page is actually about Blue Widget installation"]}

Check:

The proposed title uses the page topic and keeps one site name, which is consistent with the H1. It avoids claiming the search engine will always use the title tag verbatim.

Input:

{"pages":[{"url":"https://example.com/products/widget","type":"product","fields":{"name":"Blue Widget","author":"","date":"","price":"$19.99","location":"","hours":"","summary":"Compact replacement widget for home repairs"}}],"head_html":"<head><title>Blue Widget | Example Site</title><meta name=\"description\" content=\"Blue Widget replacement part for home repairs. Ships for $19.99.\"></head>"}

Output:

{"url":"https://example.com/products/widget","metaDescription":"Blue Widget replacement part for home repairs. Price $19.99.","status":"ok","usedFacts":["name","price","summary"],"duplicateGroup":null,"titleFindings":{"titleCount":1},"descriptionFindings":{"descriptionCount":1},"robotsFindings":{"conflicts":[]},"canonicalFindings":{"canonicalCount":0},"manualSourceChecks":["Confirm the price is current on the live page"]}

Check:

The revised description uses only supplied facts and avoids conflating price with shipping. It stays specific to the product, remains unique, and does not add unsupported claims. The output is concrete and complete, with one meta description and the existing validation fields preserved.

Input:

{"preferred_url":"https://example.com/","cluster":[{"url":"https://example.com/home","canonical":"https://example.com/home","in_sitemap":true,"internally_linked":true,"redirect_status":"200","host":"example.com","summary":"Homepage"},{"url":"https://example.com/?utm_source=newsletter","canonical":"https://example.com/?utm_source=newsletter","in_sitemap":false,"internally_linked":false,"redirect_status":"200","host":"example.com","summary":"Homepage with tracking parameter"}],"functional_parameters_to_keep":[],"copy_url":"https://news.example.org/article","canonical_url":"https://example.com/article","copy_sitemap_host":"news.example.org","canonical_sitemap_host":"example.com","sitemap_response_headers":"X-Robots-Tag: noindex"}

Output:

{"preferredUrl":"https://example.com/","canonicalChanges":[{"url":"https://example.com/home","canonical":"https://example.com/"},{"url":"https://example.com/?utm_source=newsletter","canonical":"https://example.com/"}],"sitemapKeep":["https://example.com/"],"sitemapRemove":["https://example.com/home"],"linksToRetarget":["https://example.com/home","https://example.com/?utm_source=newsletter"],"redirectCandidates":["https://example.com/home"],"removedTrackingParameters":["utm_source"],"retainedParameters":[],"copySitemapDecision":"omit copy URL from news.example.org sitemap","canonicalSitemapDecision":"include preferred URL only on example.com sitemap","relCanonical":"https://example.com/article","conflicts":["copy URL canonicalizes to another host","sitemap header noindex should be verified on live response"],"manualChecks":["Verify the preferred homepage returns 200","Verify internal links point to the preferred URL"]}

Check:

The canonicalization consolidates both duplicate home URLs to the preferred root and strips the tracking parameter from the canonical target. It also keeps the syndicated copy out of the wrong-host sitemap. The output remains factual and concrete without inventing redirects or response details not present in the input.

Input:

{"draft_paragraphs":["Blue widgets come in several sizes. Choose the right fit before installation."],"inventory":[{"url":"https://example.com/guides/widget-sizing","title":"Widget Sizing Guide","summary":"Explains how to choose the correct widget size"},{"url":"https://example.com/guides/widget-installation","title":"Widget Installation Guide","summary":"Shows how to install a widget"}],"pages":[{"url":"https://example.com/deep-page","indexable":true,"inbound_internal_links":0,"depth_from_home":4}],"source_pages":[{"url":"https://example.com/guides/widget-sizing","title":"Widget Sizing Guide","summary":"Explains how to choose the correct widget size"}],"max_links":3}

Output:

{"links":[{"paragraphIndex":1,"url":"https://example.com/guides/widget-sizing","anchor":"choose the right fit","reason":"The paragraph discusses size selection"}],"skippedCandidates":[{"url":"https://example.com/guides/widget-installation","reason":"Not chosen because only one directly relevant destination should be linked per paragraph"}],"queueItems":[{"url":"https://example.com/deep-page","priorityReason":"zero inbound internal links and depth 4"}],"validationNotes":["Verify the linked page is indexable and reachable"]}

Check:

The links are now concrete and aligned to the paragraph content. Both inventory pages are relevant, so neither is wrongly skipped. The orphan page is still queued because it has no inbound internal links and sits four clicks from home, which supports prioritization.

Input:

Example llms.txt
# Acme Docs
> Product documentation
## Guides
- [Start](/start) Quick start guide
User-agent: *
Disallow: /private

Output:

{"valid":false,"errors":[{"line":5,"message":"robots.txt directive not allowed in llms.txt: User-agent"},{"line":6,"message":"robots.txt directive not allowed in llms.txt: Disallow"}],"h1":"Acme Docs","summary":"Product documentation","sections":[{"heading":"Guides","links":[{"title":"Start","url":"/start","note":"Quick start guide"}]}]}

Check:

The markdown structure is preserved while the two robots.txt style directives are flagged as invalid. The file is correctly marked invalid, and the summary and section link are kept intact without inventing extra structure or behavior.

Input:

{"pageUrl":"https://example.com/docs/api/auth","rootStatus":{"url":"https://example.com/llms.txt","status":200,"contentType":"text/markdown"},"additionalStatuses":[{"url":"https://example.com/docs/llms.txt","status":200,"contentType":"text/markdown","scope":"Docs"}]}

Output:

{"rootFile":{"url":"https://example.com/llms.txt","status":200},"additionalFiles":[{"url":"https://example.com/docs/llms.txt","scope":"Docs","status":200}],"documentedScope":"Docs and site-wide","recommendedFile":"https://example.com/docs/llms.txt","reasoning":"The page sits under Docs and the site explicitly publishes a Docs-scoped file.","outOfScopeLinks":[],"limitations":["Discovery behavior in external clients is not guaranteed"]}

Check:

The output distinguishes the root file from the Docs-scoped file and avoids claiming any universal deepest-path precedence rule. It stays within the provided statuses and scope labels, with a cautious recommendation based on the documented Docs file.

Input:

{"links":[{"sourceAnchor":"Guides > Start","url":"https://example.com/start","status":200,"finalUrl":"https://example.com/start","approximateBytes":18000,"optional":false},{"sourceAnchor":"Optional > Archive","url":"https://example.com/old","status":301,"finalUrl":"https://example.com/new","approximateBytes":420000,"optional":true}]}

Output:

{"links":[{"sourceAnchor":"Guides > Start","finalUrl":"https://example.com/start","status":200,"approximateBytes":18000,"optional":false},{"sourceAnchor":"Optional > Archive","finalUrl":"https://example.com/new","status":301,"approximateBytes":420000,"optional":true}],"repairs":[{"sourceAnchor":"Optional > Archive","issue":"redirects to a new URL","priority":2}]}

Check:

The required link is preserved as a direct 200 response, and the optional archive link is marked as a lower-priority redirect issue. The output remains concrete and does not overstate the impact of the larger file size beyond the supplied data.

Input:

{"robotsTxt":"User-agent: Googlebot\nAllow: /llms.txt\n\nUser-agent: Google-Extended\nDisallow: /llms.txt","paths":["/llms.txt","/docs/start.md"]}

Output:

{"groups":["Googlebot","Google-Extended","*"],"matrix":[{"path":"/llms.txt","Googlebot":"allow","Google-Extended":"disallow","*":"unknown"},{"path":"/docs/start.md","Googlebot":"unknown","Google-Extended":"unknown","*":"unknown"}],"summary":"Googlebot may fetch /llms.txt, while Google-Extended is blocked by the supplied rules."}

Check:

The matrix separates Googlebot from Google-Extended and does not generalize to all crawlers. It stays strictly within the provided robots.txt rules and avoids unsupported claims about assistant behavior or universal access.

Input:

{"rawHtml":"<html><head><title>Pricing</title></head><body><div id=app></div></body></html>","renderedText":"Pricing\nPlans start at $29 per month\nIncludes email support"}

Output:

Raw document title: Pricing
Raw body text: none
Raw body word count: 0
Rendered text word count: 10
Text present only in the rendered version:
- Plans start at $29 per month
- Includes email support
Classification: thin client-rendered shell
Possible remediation: include the core pricing copy in the server-delivered HTML.

Check:

The raw body contains only an empty div, so its visible body word count is zero. The rendered text contains 10 tokens when Pricing and the two supplied lines are counted.

Input:

{"pageContent":{"questions":[{"q":"How do refunds work?","a":"Refunds take five business days."}],"editorialModel":"publisher single answer"}}

Output:

{"recommendedType":"FAQPage","rationale":"The publisher supplies one visible answer per question.","visibleQA":[{"q":"How do refunds work?","a":"Refunds take five business days."}],"jsonLdOnlyAnswers":[],"limitations":["Markup does not guarantee rich results"]}

Check:

The page has a standard question and a single publisher-provided answer, which fits FAQPage rather than QAPage. The output correctly states the answer is visible and keeps the rich result disclaimer, avoiding unsupported guarantees.

Input:

{"auditScore":82,"robotsTxt":{"googlebot":"allowed"},"headers":{"x-robots-tag":"index,follow"},"rawHtml":"<h1>Guide</h1><p>Core answer text.</p>"}

Output:

{"auditFindings":["page check score 82"],"googlebotAccess":"allowed","indexControls":["index,follow"],"snippetControls":[],"canonicalReview":"none","meaningfulTextPresent":true,"manualChecks":["confirm page stays live"],"limitations":["Passing does not guarantee AI-feature placement"]}

Check:

The output confirms ordinary eligibility signals without promising placement in any AI feature. It remains grounded in the supplied score, access rules, and visible text, while keeping the non-guarantee limitation explicit.

Input:

Illustrative data only. Client error text: OAuth discovery failed after adding https://api.timetopost.co/mcp, and the browser consent page never opened.

Output:

Illustrative assistant summary: The connection did not complete, so no account details are available. Re-add the exact endpoint https://api.timetopost.co/mcp, remove the saved server entry, and retry a fresh browser OAuth flow. Do not reuse an expired code or share secrets in chat. If discovery still fails, verify the server is reachable before attempting login again.

Check:

This is clearly labeled illustrative and does not pretend to be an exact API response. It avoids inventing account metadata or a successful login, and it recommends a fresh connection attempt rather than reusing an expired authorization code.

Input:

Synthetic example: The client currently displays only one MCP tool name: list_free_tools. The user wants to call seo_audit.

Output:

The only confirmed tool name is list_free_tools. Run that discovery tool and inspect its returned names and permission information. Do not call seo_audit unless that exact name appears in the discovery result and is available to the current session.

Check:

The answer uses only the tool name shown in the input. It does not invent discovery tools, account details, permissions, or capabilities.

Input:

Synthetic page data: URL https://example.com/article. HTML head: <title>Remote Team Content Calendar</title>. HTML body: <h1>Remote Team Content Calendar</h1><p>A shared calendar helps remote teams coordinate publication dates.</p>. No meta description is present.

Output:

Page: https://example.com/article
Title: Remote Team Content Calendar
H1: Remote Team Content Calendar
Meta description: missing
Visible supporting text: A shared calendar helps remote teams coordinate publication dates.
Recommended action: add a concise meta description based on the visible page text.

Check:

Every finding comes directly from the supplied synthetic HTML. The output does not invent a score, ranking, traffic estimate, full-site crawl result, or external tool capability.

Input:

Synthetic planning data: Platform X, timezone America/New_York, test week 2026-09-01 to 2026-09-07, metric impressions, and no account analytics or cited benchmark data.

Output:

No evidence-backed best time can be selected from the supplied data. For a concrete initial test, compare two labeled test slots: 2026-09-01 10:00 America/New_York and 2026-09-03 10:00 America/New_York. Compare impressions only after using comparable posts and observation periods.

Check:

The two times are test conditions, not claimed industry best times. The output uses the supplied impressions metric and does not invent account analytics, benchmark studies, or guaranteed performance.

Input:

Synthetic source excerpt: "Remote collaboration is easier when teams document decisions, assign owners, and record deadlines in a shared workspace." Write two X drafts and include https://example.com/article.

Output:

Draft 1: Remote collaboration is easier when teams document decisions and assign clear owners. https://example.com/article

Draft 2: A shared workspace can help distributed teams record decisions, owners, and deadlines. https://example.com/article

Check:

Both drafts are concrete and use only claims from the synthetic source excerpt. They contain the supplied URL and do not claim approval, scheduling, publication, or performance.

Input:

Illustrative data only. Draft draft_44 says: 'Launch day is here.' The intended account is acct_9, timezone is America/Los_Angeles, and the scheduler previously returned an uncertain status.

Output:

Illustrative assistant summary: read only receipt for draft_44 on acct_9 in America/Los_Angeles. Final text is "Launch day is here." The previous scheduler status is uncertain, so this receipt only records a human decision checkpoint. No scheduling, retry, or queue placement is performed here.

Check:

The output should be read only, should surface the uncertain existing job, and should not schedule, timestamp, or retry automatically.

Input:

Two user-supplied posts. Row 1: followers 12500, likes 420, comments 35, shares 10, saves 5, basis followers. Row 2: followers 12500, likes 210, comments 20, shares 6, saves 4, basis followers.

Output:

Follower-based results only: row 1 engagement total is 470, so the rate is 3.76 percent. Row 2 engagement total is 240, so the rate is 1.92 percent. The median row rate is 2.84 percent, and the aggregate follower-based rate is also 2.84 percent because 710 total engagements divided by 25000 followers equals 2.84 percent. No impression-based result is provided because no impressions were supplied.

Check:

Confirm both rows use the same follower basis, the per-row math is correct, the median for the two row rates is their average, and no unsupported impression-based claim is added.

Input:

followers 18000, price_per_thousand 20, likes 900, comments 60, shares 40, bands at least 0 but below 2 percent add 0 percent; at least 2 through 4 percent add 10 percent; above 4 percent add 20 percent.

Output:

Base quote is 360.00 USD because 18000 followers divided by 1000 equals 18, and 18 times 20 equals 360.00. Engagement total is 1000, so engagement rate is 5.56 percent because 1000 divided by 18000 times 100 equals 5.56. The matched band is above 4 percent, uplift percent is 20, and the adjusted quote is 432.00 USD.

Check:

Verify the base quote formula uses followers per 1000 times the supplied price, the engagement rate uses only the provided counts, and the band rules are explicitly user-defined.

Input:

Two records. Record 1 url https://example.com/offer?ref=home, source x, medium social, campaign spring-sale, content thread-a. Record 2 url https://example.com/offer?ref=home, source newsletter, medium email, campaign spring-sale, content hero-top.

Output:

Record 1 becomes https://example.com/offer?ref=home&utm_source=x&utm_medium=social&utm_campaign=spring-sale&utm_content=thread-a. Record 2 becomes https://example.com/offer?ref=home&utm_source=newsletter&utm_medium=email&utm_campaign=spring-sale&utm_content=hero-top. The original ref=home parameter stays intact in both URLs.

Check:

Confirm both records share the same campaign, each source and medium match the intended channel, and the original ref parameter remains unchanged.

Input:

URL https://example.com/landing?utm_source=X&utm_medium=social&utm_campaign=Spring Sale and allowed source x, allowed medium social.

Output:

Issues: utm_source uses uppercase X instead of the allowed lowercase x, and utm_campaign contains an unencoded space in Spring Sale. A corrected URL would keep the same destination path and use https://example.com/landing?utm_source=x&utm_medium=social&utm_campaign=Spring%20Sale.

Check:

Verify the report flags case sensitivity and encoding problems, and that the corrected URL preserves the destination path and does not invent any other changes.

Input:

Planned URL https://example.com/page?utm_source=x&utm_medium=social&utm_campaign=spring and landed URL https://example.com/page?utm_medium=social&utm_campaign=spring.

Output:

Lost parameter: utm_source. Kept parameters: utm_medium and utm_campaign. The planned URL already contains utm_source=x, so readding it would not fix the loss; the redirect or intermediary must be corrected to preserve that parameter, and the original planned URL must then be retested.

Check:

Confirm the lost parameter is identified from the planned URL and not treated as present just because other UTM fields remain.

Input:

Goal book. Links are homepage, booking page, pricing page, and Instagram DM. Booking page reaches the goal in one step, pricing page supports the goal but is not the final action, homepage is broad, and DM is secondary.

Output:

Primary link: booking page. Secondary links: pricing page and Instagram DM. Overflow link: homepage. Profile sentence: Book a 20-minute consultation.

Check:

Confirm only one link is primary, the shortest path to the goal wins, and the profile sentence clearly names the main action without adding extra promises.

Input:

Synthetic source excerpt: "In our internal workflow review, teams sent a short email summary every Friday. The summary listed open tasks, assigned owners, and due dates. Team members reported that the list helped them identify pending follow-ups sooner." Request a 3-post X thread, one caption, one verbatim pull quote, and one email teaser.

Output:

Thread post 1: A short Friday email summary can list open tasks, assigned owners, and due dates.

Thread post 2: In the workflow review, team members reported that this list helped them identify pending follow-ups sooner.

Thread post 3: Practical format: send one weekly summary covering each open task, its owner, and its due date.

Caption: A simple Friday summary can keep open tasks, owners, and due dates visible to the team.

Pull quote: "Team members reported that the list helped them identify pending follow-ups sooner."

Email teaser: What should a weekly follow-up summary contain? This article describes a Friday format covering open tasks, assigned owners, and due dates.

Check:

The thread has exactly three posts. Every factual statement is grounded in the full synthetic excerpt, and the pull quote reproduces one supplied sentence verbatim.

Input:

spend 500, currency USD, impressions 100000, clicks 2000, conversions 50.

Output:

CPM is 5.00 USD because 500 divided by 100000 times 1000 equals 5.00. CPC is 0.25 USD because 500 divided by 2000 equals 0.25. CPA is 10.00 USD because 500 divided by 50 equals 10.00. CTR is 2.00 percent because 2000 divided by 100000 times 100 equals 2.00.

Check:

Verify each formula uses the correct denominator and that all four rates are included with no benchmark or comparison language.