feat(marketing-skill): add ai-decision-authority

[kubersharmax]

Adds a skill for classifying which AI outputs require human approval, using a three-zone model sorted by reversibility, exposure and frequency.

Summary

marketing-skill/skills/ai-decision-authority/SKILL.md

Marketing teams adopting AI tend to land in one of two failure modes: review every output, which collapses adoption because the tool becomes slower than doing the work by hand, or review nothing, which accumulates compliance exceptions quietly. Both come from applying one governance rule to every decision.

This skill sorts decisions into three zones by stakes rather than seniority, and gives the agent a rule per zone plus the cost of misclassifying in either direction. It handles a single decision or a batch audit of many workflows.

I checked the 48 existing skills in marketing-skill/skills/ before building this. Nothing currently covers AI governance, human-in-the-loop policy, or decision authority, so there is no overlap.

One useful distinction it encodes: most workflows span zones, so govern the boundary rather than the whole workflow at its highest zone. "Drafts the newsletter and schedules it" is Zone 1 drafting plus Zone 2 sending, and gating only the send is where most of the available adoption is won.

Checklist

• Target branch is dev (not main — PRs to main will be auto-closed)
• Skill has SKILL.md with valid YAML frontmatter (name, description)
• Scripts (if any) run with --help without errors — no scripts in this skill
• No hardcoded API keys, tokens, or secrets
• No vendor-locked dependencies without open-source fallback
• Follows existing directory structure (domain/skill-name/SKILL.md)

Note on frontmatter: this template's checklist mentions a license field, but CONVENTIONS.md section 2 states only name and description are allowed and that PRs with extra frontmatter fields will be rejected. I followed CONVENTIONS.md. Happy to add license if the template is the current rule.

Also followed: under 500 lines (114), opinionated rather than a list of options, Anti-Patterns section included, and Cross-References to marketing-ops, marketing-strategy-pmm, launch-strategy, brand-guidelines and content-production. No plugin.json, since CONVENTIONS notes skills roll up into the domain plugin automatically. No files touched outside the new directory, so the skill count and auto-generated manifests are unaffected.

Type of Change

• New skill
• Improvement to existing skill
• Bug fix
• Documentation
• Infrastructure / CI

Testing

No automated tests: this skill is instructions only, with no scripts, so there is nothing to run --help against.

Verification was manual. I walked a mixed set of marketing decisions through the three questions to check the zone boundaries produce sensible results: content variants and internal competitive summaries fall in Zone 1; campaign briefs, partner communications and a newsletter send fall in Zone 2; pricing announcements and regulatory claims fall in Zone 3. I also checked the split case, where drafting and sending inside one workflow separate across two zones rather than collapsing to the higher one.

The three questions are order-dependent by design. Asking about frequency before reversibility gives wrong answers on low-volume reversible work, so the skill instructs the agent to ask reversibility first.

The underlying model comes from applied use in an enterprise product marketing function rather than from a benchmark, and the SKILL.md says so.

Provenance

Based on the Augmented Marketing Decision Architecture, from an observational deployment across 16 product launches in one product marketing function. The SKILL.md states the scope and limits of that evidence rather than presenting the numbers as universal benchmarks.