Google Search Central ties helpful pages to clear intent and substantive depth, both decided in the brief, not patched in edit. Teams report that expensive agent draft rework usually traces to briefs with vague ICP, missing evidence links, or keyword-stuffed outlines, not to the writing model. A content brief skill encodes approval judgment before any model writes prose.
TL;DR
- The skill validates audience, angle, evidence, links, and range, not blank section headings.
- Use the Content brief skill anatomy: inputs, rules, examples, eval hooks.
- Parameterize funnel stage, risk tier, primary keyword, and persona.
- Trigger downstream AI content pipelines only after brief QA passes.
- Version the skill with version marketing workflows.
Start with one seed brief
Pick one seed brief. Validate JSON against the skill. Fix link and JTBD failures. Then trigger research and draft. Skipping validation sends bad inputs straight to expensive agent runs.
Anthropic skills documentation aligns with treating brief validation as a tagged skill file, not a checklist PDF.
What a content brief skill governs
The skill sits upstream of draft agents. It accepts a brief JSON or form payload and returns pass, fail, or structured fix requests. It does not replace human strategists on net-new pillars, it stops agents from executing garbage inputs.
| Stage artifact | Without skill | With governed brief skill |
|---|---|---|
| Audience | Vague persona label | ICP + job + stage |
| Evidence plan | “Add stats” | Named sources required |
| Internal links | Placeholder slugs | Live slug allowlist only |
| Intent | Keyword only | intent_gaps[] covered in outline |
| Range | Missing | min/max word targets |
Extraction guidance from turn marketing expertise into ai skill applies: interview editors on where they send briefs back.
Content brief skill anatomy
The Content brief skill anatomy mirrors other production skills in marketing agent skills programs.
What inputs does a content brief skill require?
Typical keys: `primary_keyword`, `target_query`, `persona`, `funnel_stage`, `risk_tier`, `word_count_min`, `word_count_max`, `internal_link_targets[]`, `intent_gaps[]`, `outline_h2[]`. Schema-validate at intake.
Which rules do brief reviewers enforce?
Rules must be falsifiable. Examples: opening must plan one cited stat; internal links must be subset of live registry; outline must map each intent gap to an H2 or integrated section, not a bolt-on “SERP misses” H2 unless brief explicitly allows.
What examples belong in brief skill fixtures?
Ship pass/fail pairs: thin keyword-stuff brief vs brief with persona JTBD and evidence rows.
How do eval hooks catch bad brief payloads?
Declare pass threshold and tie to marketing skill evaluation golden files.
| Block | Owner | Change trigger |
|---|---|---|
| Inputs schema | Content ops | New content type |
| Rules | Senior editor | Brand refresh |
| Examples | SEO lead | Algorithm update |
| Eval hooks | Marketing ops | Rubric change |
Rules brief reviewers actually enforce
These rules belong in most production brief skill files. Tune weights by risk tier per the brief skill rule checklist ops publishes beside the tag.
- Persona JTBD: `target_persona.jtbd` must be a complete sentence with role and outcome.
- Evidence row: At least two planned external sources with URL or named publisher before draft trigger.
- Link health: Every planned internal link resolves to a published slug.
- Outline integrity: Each `intent_gaps[]` item appears in `outline` key points, not only in free text notes.
- Range sanity: `word_count_max` ≥ `word_count_min` and within ops band for content type.
- Keyword fit: Primary keyword appears in planned title or H2, not stuffed in every line.
High-risk tiers require human sign-off on the skill output even when score passes auto threshold.
Worked failure: draft slug in brief
A brief listed `/blog/upcoming-launch` as an internal target. The content brief skill blocked the draft run on link health, the unpublished slug never reached the draft agent. Fix: swap to a live URL or remove until publish. This first-hand pattern matches ops reports: internal link errors are cheaper to catch in brief QA than in CMS rollback.
Golden tests for brief skills
Maintain 10, 15 brief payloads: TOFU explainer, MID workflow guide, BOFU comparison support, listicle, refresh. Expect pass or named fail codes.
| Case | Expected |
|---|---|
| Complete brief JSON | Pass |
| Missing persona JTBD | Fail |
| Draft slug in links | Fail |
| intent_gaps not in outline | Fail |
| word max < min | Fail |
Regression-test on every skill tag promotion. Log scores beside tag in ops registry.
Sample brief payload fields
| Field | Example | Rule tested |
|---|---|---|
| target_persona.jtbd | “Ops lead needs brief gates before agents draft” | JTBD completeness |
| intent_gaps[] | “No SERP template for eval hooks” | Outline mapping |
| internal_link_targets | `/blog/ai-content-pipelines` | Live slug only |
| word_count_min / max | 1600 / 2400 | Range sanity |
| risk_tier | high | Human gate flag |
Authors fill the form once. The skill returns machine-readable errors editors can fix without interpreting model prose.
Maintenance and ownership
Assign a named owner for the brief skill tag. Review rules when brand voice, legal, or CMS slug registries change. Patch for typo-level rule edits; minor for new optional fields; major when pass or fail behavior shifts.
Run the golden brief set monthly even without code changes, slug registries drift as new posts go live. Document results in the same changelog you use for workflow pin promotions.
Editors who trust agent output usually fix judgment upstream first. The governed brief skill is that upstream gate.
Wire the skill into content pipelines
In reusable marketing workflows brief-to-publish flows, invoke the skill after brief authoring and before research/draft stages.
``` Brief form → brief skill → (pass) research agent → draft agent → QA ```
Pin skill version in workflow manifest. Co-promote with eval rubric when rules change.
Editors who trust agent output usually fix judgment upstream first, but that only sticks when the validator lives in version control instead of a one-off doc. The governed brief skill is that upstream gate, encode outline, evidence, and link rules once, then run every brief through the same pass/fail contract before research or draft agents spend tokens.
When brief quality compounds, outline and evidence travel as skills and context instead of dying in Notion comments. Metaflow lets teams harden the validator once, pin the tag, and attach it to every content workflow so discovery tweaks become durable defaults, not disposable chat fixes.
Frequently Asked Questions About Building a Content Brief Skill
What is a content brief skill?
It is a versioned skill file that validates brief payloads against audience, evidence, link, outline, and range rules before draft agents run. It returns pass/fail and fix codes instead of silent acceptance.
How is a content brief skill different from a template?
Templates show headings; the skill enforces testable rules on JSON or form fields. Templates do not block bad links or missing JTBD. Metaflow skills store those rules outside templates so workflows call the same validator by version tag.
What fields belong in a content brief skill?
Inputs schema, rule set, pass/fail examples, eval threshold, and optional human-gate flags by risk tier, not long prose instructions alone. Metaflow parameterizes funnel stage and risk tier at runtime on the same skill file.
How do you test a content brief skill?
Golden brief JSON fixtures with expected outcomes; rerun on every promotion. Metaflow eval jobs can execute those fixtures against candidate tags before you update the production pin.
Where does a content brief skill sit in a workflow?
Immediately after brief creation and before research or draft stages in content pipelines. Failed briefs loop to authors with structured errors. Metaflow flows can branch on fail codes to human queues without starting expensive draft runs.
