A writer-ready brief with one primary topic, audience and intent, a supported angle, source-backed claims, outline, exclusions and measurable acceptance criteria.
Validate a content topic with current search-demand signals and cited public sources, then produce a writer-ready brief with a supported angle, outline and measurable acceptance criteria.
What this Skill does—and does not do
The Skill does not write or publish the finished article, reproduce source text, invent statistics, present vendor claims as independent proof or expand research without approval.
Inputs
A seed topic or business question
Audience, business objective and desired action
Country, language, exclusions and any required first-party sources
Outputs
Primary topic and supporting-topic evidence
Cited claim ledger with source roles and cautions
Writer-ready brief with angle, outline, exclusions and acceptance criteria
Reusable brief artifact and complete research receipt
How this Skill works
1
Share the starting information
Give your agent the inputs listed above and explain the result you need.
2
Review the plan
Your agent reviews the proposed tools, inputs and maximum cost before any paid work begins.
3
Let your agent run the workflow
After you approve the plan, your agent uses the live Scrollport tools shown on this page and keeps the work inside the agreed limits.
4
Review the finished result
After approval, your agent returns the promised output and reports the tools used, final cost and any limitations.
Full instructions your agent receives
Use get_run({ run_id, wait_seconds? }) to read a saved run before any retry; no idempotency key is accepted. Waiting defaults to 50 seconds and accepts 0–120. run_tool starts only and requires tool_id, input and one UUID per paid intent; retain that UUID for an exact retry after an uncertain start. Every state retains run_id.
Evidence-led Content Brief
Turn a seed topic into an evidence package a professional writer can use. The workflow combines current keyword signals with source research, then makes the agent show which claims are supported, which are vendor perspectives and which are editorial recommendations.
Use one authorized Scrollport connection and its search_tools, inspect_tool, run_tool, get_run, list_apps and get_wallet control tools. Never call a supplier directly.
Before starting
Use any supplied product, positioning, audience, editorial or client brief before asking questions. If .agents/product-marketing.md exists, read it as context; confirm the task-specific audience, intended action, market and excluded claims rather than treating the file as current evidence.
If the user has only a domain or a broad goal and has not selected a topic, route to seo-opportunity-map
Cost and approvals
How pricing works
Inspected keyword-batch and web-search prices for one bounded validation and evidence pass.
Your approval points
Approve the topic set, market, research query, requested result count, exact paid inputs and maximum total before the first metered run
Approve any second search, broader market, larger topic batch or additional provider
first. This Skill turns one chosen opportunity into a production brief; it does not decide the site's complete content strategy.
Outcome boundary
Produce a writer-ready brief containing:
a defined audience, job and intended action;
one primary topic and a small supporting-topic set;
current demand, difficulty, intent and SERP evidence when available;
a differentiated angle grounded in cited public sources;
a claim ledger, outline, exclusions and acceptance criteria.
This Skill does not write the finished article or publish anything. It does not copy source prose, manufacture a statistic, or turn a vendor's assertion into independent fact. Research supports the brief; it does not replace editorial judgement or first-party expertise.
Inputs and trial-safe defaults
Require:
a seed topic or question;
intended audience and business objective;
the action the finished content should enable;
country and language;
excluded claims, products, competitors or sources;
any required first-party material supplied by the human.
Default to at most five unique topics in one keyword batch and at most five public sources in one research call. Stop if the inspected plan exceeds $0.050000. Do not run a second search simply because the first sources challenge the preferred angle.
Select and inspect the route
Discover by intent and inspect every selected tool immediately before use. The currently verified route is:
dataforseo.keyword-overview for topic demand, difficulty, intent and SERP context;
exa.search for current public evidence and source diversity.
If either required tool is not live, its schema cannot express the bounded plan, or its current unit price cannot be mapped to a maximum, stop. Do not quietly replace structured search data with a general web result.
Save the exact market, deduplicated topic batch, research query, requested result count, price unit, inspected price and maximum for both calls.
Plan and approval
Construct a narrow topic set locally from the supplied seed and objective. Do not use a paid expansion call in trial-safe mode. Present:
the primary question and intended reader;
at most five topics to validate;
one research query and at most five requested sources;
any required domains or excluded domains;
exact tool ids, paid inputs, inspected prices and maximum total.
Wait for explicit approval. Changing the audience, market, topics, query, result count or provider invalidates it.
Do not put credentials, approval URLs or licensed full-text source copies in state.
Execute
1. Validate the topic set
Run one approved dataforseo.keyword-overview batch. Preserve only fields the provider returned. For each topic record volume, difficulty, intent, competition and SERP features when present.
Select a primary topic using business fit and search intent first, then demand and difficulty. A lower-volume, high-fit commercial topic can be a better primary target than a broad informational phrase. Explain the choice; do not invent a composite provider score.
2. Gather current source evidence
Run the approved exa.search request once. Inspect each returned result before using it. Prefer a mix of primary or expert sources, recognised practitioner analysis and concrete examples. Record title, publisher, URL, date when returned, source role and the exact claim it can support.
Treat vendor research, consultancy viewpoints and practitioner essays as such. Do not repeat an embedded statistic unless the result exposes a credible original source and the claim is relevant. A relevant snippet is a lead, not permission to fabricate details beyond it.
3. Build a claim ledger
Before the outline, create a compact ledger:
supported claim;
source URL and publisher;
evidence type: primary, expert analysis, vendor example or practitioner perspective;
caveat or counterpoint;
planned section.
Exclude claims that lack enough source evidence. Identify assertions that must come from the human's first-party experience rather than public research.
4. Design the brief
Create the editorial angle locally. The outline should answer the reader's job, not merely place keywords in headings. Map each substantive section to at least one ledger item or explicitly label it as first-party opinion to be supplied.
Include:
working title and one-sentence promise;
reader, problem, objective and intended action;
primary topic, supporting topics and intent rationale;
differentiated angle and what the piece will not claim;
section-by-section outline with evidence notes;
source and claim ledger;
original examples, screenshots or first-party inputs still needed;
internal-link and call-to-action suggestions when supplied context supports them;
completion checklist for the writer and editor.
Use the output template or an equivalent structure so the brief is directly reusable outside the conversation.
Deliverable and acceptance
Return the brief plus exact tool ids, run ids, final costs, total spend and any skipped optional work. Distinguish DataForSEO fields, Exa-returned source evidence and agent editorial judgement.
On resume, poll every saved non-terminal run id before creating a new one. Reuse successful results. If a server estimate requests confirmation, hand the approval URL to the human and wait. Do not repeat a paid search because a local draft was lost.
The Skill passes only when:
one primary topic is selected with an explicit market and intent rationale;
every externally verifiable substantive claim has a usable citation;
vendor claims and statistics are labelled or excluded;
the outline is specific enough for a writer to draft without repeating the research pass;
the brief stays inside the approved calls and cost;
exact run ids and final spend are recorded.
A list of keywords and links without a coherent, supported editorial plan does not qualify.
Quality checks
Before delivery, confirm that:
the angle solves the reader's job rather than restating the keyword;
the primary topic was chosen for audience and intent fit before volume;
every substantive external claim maps to a usable source and source role;
requested first-party expertise is explicit rather than fabricated; and
the outline, evidence ledger and acceptance criteria can be handed to a writer without repeating the research pass.