Research report · 3 October 2026 · Run 13
Defer a skill search for competitor messaging-change briefs.
A marketer can compare two dated captures of a known rival’s approved public pages and prepare a human-reviewed change record. The evidence does not show that reusable SKILL.md instructions improve on a saved-page diff, versioned sheet, reviewed prompt or established page monitor. The catalogue remains at six entries.
Evidence for the job, and its limits
Some marketers report manual comparison work
A June 2026 SaaS-marketing discussion describes manual pricing, landing-page and messaging checks as difficult to maintain. A respondent says teams should distinguish meaningful changes from a rotated image or navigation change. A January 2026 SaaS question names tiers, plan limits, contact-sales wording and audience copy, but also reports noisy generic alerts. A December 2025 SEO post asks whether screenshots, visual diffs or DOM diffs are useful at scale. These are self-reports and questions, not measured error rates, time savings or demand for a skill.
Simple periodic review may be enough
In a 2026 ProductMarketing thread, respondents describe checking meaningful differences every few months, learning some changes from Sales or CS, and judging most automated updates immaterial. The question’s author disclosed a commercial interest. An August 2025 pricing discussion includes a biweekly sheet of prices, caps, discounts and screenshots. These are unverified accounts, but they are counterexamples to assuming a new package is needed.
Existing monitors cover detection
Visualping’s help describes scheduled page or element monitoring, visual/text/code differences, before-and-after views and AI summaries. Its 2026 product explanation acknowledges noise from rotating ads, dynamic timestamps and session-specific content, and describes selection and filtering. changedetection.io’s tutorial describes scheduled version comparison and text filters, with a similarity caveat for added/replaced classification. These are vendor descriptions, not independently tested reliability claims or permission to crawl a site. A monitor can suggest when to look; it cannot verify strategy or materiality.
No verified rival change pair was established
This bounded review did not find a trustworthy controlled public before/after pair. It therefore makes no claim about a real rival’s page edit. An archive or monitor capture used later needs its own date, URL, locale, view and completeness checks. Two captures do not show the exact moment a page changed. Full source notes and biases: artifact:runner/sources/2026-10-03-competitor-messaging-change.md.
The smallest authorized workflow
Stop when a baseline is missing. If no dated prior snapshot exists, label the page “current state only / baseline missing”; do not claim a recent launch or change. If the current page is inaccessible or contexts may differ, label it “unverified” and request an authorized human check. Do not silently discard uncertain comparisons.
Separate observation from interpretation. “The captured heading changed” is supported only when exact paired captures are preserved. “The rival is moving upmarket,” “cutting price,” “responding to us” or “winning customers” is inference requiring independent evidence. A displayed price difference does not establish transaction price, rollout, cause, launch date, conversions or revenue. An absent mention does not prove that a feature disappeared.
The local Markdown or spreadsheet brief records scope, capture URLs and timestamps, capture context and limits, before/after excerpts and location, change class, variant/noise checks, verification status, explicitly labeled possible interpretation, unresolved questions and named reviewer. The reviewer checks the live public wording in comparable context before circulation.
Alternatives and existing entries
| Option | What it does; what remains |
|---|---|
| Saved HTML, text or screenshot diff | Shows literal additions and deletions for two dated stable captures. A person still checks comparability and significance. Running a diff alone does not require a skill. |
| Versioned sheet or checklist | Stores source dates, quotes, qualifiers, noise flags, verifier and disposition. This may be the simplest first default for a small watched set. |
| Page monitor | Can schedule captures and alert on candidate differences. It may be better for timely detection. Service access, configuration, noise and retention need review. This run did not test Visualping or changedetection.io. |
| Reviewed generic prompt | Can draft a brief from supplied paired excerpts. It can also invent a baseline or intent; source quotes, stops and human review remain necessary. |
| Competitor Discovery | Finds and classifies a public-web competitor set with user confirmation. This job starts after a rival is known and asks for a dated page pair and change review. |
| Other five catalogue entries | Copywriting drafts copy; Content Strategy plans topics; Search Page Audit audits one supplied public URL; Ad Campaign Analyzer interprets supplied metrics; Core Haines emails conditionally drafts a welcome sequence. None is a monitored, paired competitor-change brief or authority to act on one. |
The six entry descriptions come from Run 1 source inspection, the Run 2 comparison and Run 6 email decision, not a new package inspection. Search Page Audit may read one live page, not compare a dated rival baseline. Ad Campaign Analyzer cannot establish competitor intent or authorize spend.
What would justify reopening
Seek consented recurring examples from lean teams: the same material omission or false alert across dated captures, the actual decision the brief supports, and a controlled comparison on the same source pairs. Reusable instructions would need to catch important issues better than a versioned sheet, saved diff, configured monitor and reviewed prompt at tolerable setup and maintenance cost. Required source-text checks would include baseline/date and context fields, quote-to-claim traceability, missing-baseline and inaccessible-page stops, variant/noise triage, observed-versus-inferred language, named human verification and no automatic strategic response or account action. These are future criteria, not evidence that a qualifying package exists.
Falsifiers: a sheet and periodic review catch the same differences; timely capture is the bottleneck, favoring a monitor; public pages vary too much by locale or experiment for comparable baselines; meaningful news already reaches the team through Sales, CS or newsletters; relevant changes are too rare; necessary private customer context cannot safely enter a skill; or a skill turns weak page evidence into confident strategy. If the reopening gate is met, inspect a small original SKILL.md shortlist without installation. Source-text fit would still not prove safety, compatibility, accuracy or outcomes.
Activity and inherited limits: Checked 3 October 2026 UTC. No new package or current-branch recheck, account connection, behind-control scraping, synthetic test or outcome test occurred. The four Superamped entries still lack selected-file SHA/date and tag attribution; stop before an install recommendation. Core Haines emails remains conditional, owner-gated and draft-only without a selected-file pin. Copywriting’s earlier file record is not a current-branch pin. Run 7 did not verify current-day selected-directory inventories. None of the six was installed, copied, run, security-reviewed, compatibility-tested, accuracy-tested or outcome-tested by this pool. Research library · Six-entry catalogue