trend-linkedin — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited trend-linkedin (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Collect evidence of what relevant LinkedIn audiences are discussing and how posts deliver useful substance. Return source-linked observations and bounded opportunities; drafting and publishing are downstream steps. Preserve the distinction between a public observation, a study result, and a craft inference.
LinkedIn uses profile information and interaction history in retrieval and ranking; it publishes no universal comment multiplier or single dominant signal (official, LinkedIn Feed Engineering, 2026-03). Public engagement cannot reveal dwell time, private conversions, or why a ranker distributed a post. Don't infer those from length or paragraph rhythm (directional).
Keep personal profiles separate from Company Pages. Metricool found similar impressions per post, but profile engagement averaged 2.60% versus 1.60% for Pages (measured, Metricool, n=673,658 posts from 63,108 accounts, 2026-04). Account type belongs in every comparison.
Record the audience, professional problem, campaign stage if supplied, language, target geography/timezone, and observation window. Include practical alternatives to the favored product or thesis. Distinguish exact quotes from paraphrases and redact private material (directional).
Start from the signed-in interface and follow controls actually present. The entry URL below is a navigation starting point, not a guaranteed harvest API; record redirects and unavailable controls (directional). LinkedIn's User Agreement bars scraping and unauthorized automated access (official, LinkedIn User Agreement, effective 2025-11-03): keep collection to a small, owner-authorized reading sample at human pace, cite posts by URL, and don't bulk-copy profiles or contact data.
| Surface | Entry and procedure | What to record |
|---|---|---|
| Personalized feed | Open https://www.linkedin.com/feed/; confirm the loaded surface before scanning | Signed-in context, visible filters, suggested versus followed sources where labeled |
| Topic search | Use the current search box, select Posts if offered, then inspect available sort/date filters | Query, displayed filter names, resulting URL and timestamp; never assume a date filter was applied |
| Creator or expert activity | Open a discovered profile and follow its visible Activity/Posts controls | Actual activity URL, time span inspected, visible account context |
| Company Page | Follow the current Page's posts, event, or publication controls | Page identity, original versus employee/customer contribution, format |
| Article / Newsletter / Event | Follow a visible source link from a relevant account or post | Actual URL, publication/event date, authorship, availability restrictions |
Don't construct hashtag-feed URLs or hard-coded sortBy parameters. If a deep link fails, return to the current UI and record the failure; don't silently substitute a different query. Treat hashtags as searchable text only when the current interface supports the operation (directional).
Scan independent sources where available; scroll until additional items stop changing the evidence or the requested scope is met. Avoid a fixed scroll quota. If login, pagination, or access controls block the sample, return a partial brief with the boundary stated (directional).
For each candidate, record the canonical post URL; author/account type; posted time and observation time; visible follower context; format; link placement; and public reaction, comment, and repost counts. Use “not visible” for inaccessible fields. Inspect the actual asset and a relevant slice of replies; label that slice's scope (directional).
Capture the opening exactly within quotation limits and note what was visible before expansion on the observed client. Don't assign a fixed mobile character fold. Record total length only if the full text was accessible; it doesn't prove dwell time (directional).
Inspect buyer substance: the problem being discussed, proof provided, stated conditions, credible objections, and the requested action. Separate author replies from independent participants where visible. A comment asking about integration, applicability, or access can indicate evaluation; it isn't a verified lead (directional).
Record hashtags, relevant mentions, collaboration labels, sponsorship, and visible Content Credentials without inferring hidden status. Credentialed media can display a provenance icon; absent icons don't establish human origin (official, LinkedIn Credentials Help, 2026-09 review, undated).
For a launch-focused brief, extend the evidence record with this compact handoff (directional). Capture observed material only; mark missing fields unknown.
| Harvest field | Record and boundary |
|---|---|
| Campaign context and shape | Supplied audience, buying situation, promise, and stage; observed announcement, demo, customer proof, founder decision, or objection answer. Don't infer an unseen campaign plan. |
| Bounded proof | Source URL, artifact, conditions, outcome period, denominator, exclusions, and competing explanations where supplied; distinguish author claims from verified results. |
| Action and participation | Actual link placement, CTA or absence, disclosure, and founder-versus-Page context; record evidence of voluntary community participation, leaving consent unknown when unstated. |
| Launch-week observations | Build a source-linked table of actual post dates, observed shapes, replies, and available outcomes; state the inspected window and unknowns. Don't convert gaps into a prescribed schedule. |
| Dimension | Prefer | Preserve limitation |
|---|---|---|
| Audience fit | A clearly relevant buying situation and identifiable professional context | Titles and badges don't validate expertise |
| Evidence quality | Inspectable work, source-linked outcomes, conditions and counterexamples | A claim in a post remains the author's claim |
| Conversation | Independent questions or reasoned disagreement that affect a decision | Raw comment totals may include author replies |
| Relative response | Comparable account type, format, audience size, post age, and visible history | Public counts don't expose impressions or causal lift |
| Reusable pattern | A task, proof structure, or unanswered question that fits the brief | Don't copy the author's lived experience or sentence template |
The rubric is directional. Use “prioritize,” “watch,” or “insufficient evidence,” with a reason. A comment-to-reaction ratio may describe the observed sample, but supplies no breakout threshold or ranking weight. Leave it undefined when reactions are absent. Don't divide by an invented follower count, infer a reach rate from reactions, or label reaction diversity an algorithmic signal.
Metricool observed stronger results for some document and multi-image formats than for video in its cohort (measured, Metricool, n=673,658 posts, 2026-04); AuthoredUp observed stronger document than video results in its personal-profile cohort (measured, AuthoredUp Formats, n=3M+ personal-profile posts, 2026-09). Capture relevant examples of these formats as testing opportunities, without inferring a universal video advantage (directional). Capture what the asset demonstrates. Don't prescribe slide counts or a video-duration optimum from format alone (directional).
Polls aren't categorically dead: Socialinsider reports 4.50% engagement in its Q2 table, but its methodology dates conflict with that period (measured, Socialinsider, n=1.3M business posts, 2026-03). Keep that caveat with the claim; prioritize the account's observed outcomes over a cross-vendor leaderboard (directional).
Standard posts allow 3,000 characters; Articles are a separate long-form feature (official, LinkedIn Post Help, 2026-08). Never call a long feed post an Article. Record Newsletter or Event formats only when the source identifies them; don't infer current Audio Events availability from old content (directional).
Link-post impressions/interactions differed by account type: profiles −27%/−20%, Pages +51%/+41% (measured, Metricool, n=673,658 posts, 2026-04). This doesn't test whether a first-comment link is safer. Describe link placement without assigning a penalty (directional).
Half of impressions arrived within 48 hours in Metricool's cohort, not a hard feed expiry (measured, Metricool, n=673,658 posts, 2026-04). Replying correlated with about 30% higher engagement (measured, Buffer Replies, n=72,000 LinkedIn posts, 2025-12). Record observed timestamps; don't infer a required reply count or minute cutoff.
If timing is requested, offer audience-timezone tests. Buffer found a 3–8 PM window, with Wednesday 4 PM and Friday 3–4 PM strongest (measured, Buffer Timing, n=4.8M+ posts, 2026-07). Rotate global-audience tests and compare like-for-like material (directional). Don't impose a daily maximum: Buffer's LinkedIn frequency analysis found gains at higher cadence (measured, Buffer Timing, n=2M posts, 2026-07).
Creator Mode's toggle was removed in March 2024 (official, LinkedIn Creator Mode Help, 2024-03). Verified-member filtering exists in feed conversations and comments (official, LinkedIn Authenticity, 2026-06). Record any active filter as sampling bias; assign no Premium, verification, or Top Voice reach multiplier.
Relevance-ranked search and a signed-in feed aren't a platform census. Report surfaces and filters, plus inaccessible material. Compare repeated angles across unrelated sources before calling an idea saturated; don't maintain a static dead-take list (directional).
LinkedIn acts on scaled automated comments and restatement-only replies, and reduces wider distribution for apparently AI-generated posts without perspective (official, LinkedIn Authenticity, 2026-06). Flag observable duplication or empty discussion without declaring authorship from style. Exclude suspicious activity from recommendations only with a stated observation and uncertainty (directional).
Don't borrow another platform's engagement definition. Public counts omit useful LinkedIn clicks and media activity (measured, Metricool Press, n=673,658 posts, 2026-04). Request authorized analytics if the brief requires qualified traffic or conversion claims; otherwise mark those outcomes unavailable.
Illustrative schema; every bracketed field requires source capture:
Source: [post URL], observed [timestamp/timezone]. Author: [name], personal profile. Published: [source time]. Format: native document. Public counts: [reactions], [comments], [reposts].
>
The opening names an invoice-review problem. The document shows assignment rules; the inspected replies ask about exceptions. The author explains the manual fallback. The source gives no verified conversion result. Prioritize the unresolved exception question for the brief; don't infer dwell time from the document format.
Why it works: the comparison rubric prioritizes buyer relevance and inspectable substance. The note separates observed conversation from unavailable business outcomes.
Weak, illustrative: “Carousels are winning; use more of them.”
In [observed sample], document posts explained approval workflows, while multi-image posts showed completed installations. These serve different reader tasks. [Source URLs] support the distinction. The sample came from [surface and filters], so it doesn't establish a platform-wide format shift.
Why it works: the harvest procedure preserves asset purpose and sample scope. The brief supplies a writing opportunity without copying a hook or claiming an algorithm preference.
Illustrative source-placeholder comparison; fill each field from the inspected sources:
[Mechanic profile post URL] explains an evening repair appointment; [workshop Page post URL] shows eligible repairs. Record [post dates], [observation time], [formats], and [visible counts] separately. Compare each with [its own comparable history URLs]; keep impressions and bookings unavailable unless authorized analytics supplies them. The profile has more visible comments in this scenario, but account context and post age differ.
Decision: retain appointment eligibility as a topic opportunity. Don't rank profile versus Page effectiveness from unmatched public totals (directional).
Illustrative source-placeholder rejection:
Proposed trend: repair demos are producing more bookings. At [candidate URL], [observed timestamp/timezone], the opening mentions appointments, but [access restriction] prevents inspection of the full asset and replies. Public counts: [visible fields or not visible]. Impressions, clicks, and bookings: unavailable. Search scope: [query, surface, filters, inspected window]; further results blocked by [observed limit].
Decision: return a partial brief and mark the proposed booking trend insufficient evidence. The accessible opening supports only a topic observation; neither the demo's proof nor its business outcome can be assessed (directional).
Re-validate when search/filter URLs or feature names change; feed engineering or disclosure policies update; new vendor reports revise cohorts or findings. Validated: 2026-10
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.