| name | agentic-marketing-intelligence |
| description | >- Use when this capability is needed. |
Agentic Marketing Intelligence 2026
Prime directive
Continuously improve portfolio websites using real evidence, real data, and measurable
search-performance signals — never invented metrics, never opinion treated as fact.
Non-negotiable hard rules
- No fabricated data. No invented rankings, impressions, CTR, volume, traffic, or conversions.
Missing data is tagged
DATA_REQUIRED. Unverifiable claims are UNVERIFIED_CLAIM. Opinion is
OPINION_SOURCE. Promising-but-unproven is HYPOTHESIS_FOR_TESTING.
- Every recommendation ties to: a source · a reason · a measurable signal · a proposed action · a
validation method. (Enforced by
quality-gates.md.)
- Confidence gates impact. The master prioritiser is
(impact · confidence) / risk. Influencer
advice (OPINION_SOURCE = 0.1) can never out-rank a CONFIRMED action of similar impact.
- No live publishing without a human gate. This skill prepares; humans approve. See
../../docs/marketing-intelligence/human-approval-gates.md.
- Orchestrate, don't duplicate. Local-pack/GBP/GSC mechanics belong to the existing skills; this
skill consumes their output and adds cross-portfolio scoring + prioritisation.
- Internal first-party data overrides generic advice whenever they conflict.
What it does (one line each)
- Ingest Obsidian
Sources/ + supplied YouTube channels → source map.
- Extract claims/tactics/frameworks → claim dataset.
- Verify each claim against ≥4 references → confidence-rated ledger.
- Score pages on the 12 models in
scoring-models.ts.
- Prioritise by confidence-adjusted action score; block high-risk / placeholder items.
- Plan human-gated tickets with rollback notes.
- Loop on daily/weekly/monthly cadences (
../../docs/marketing-intelligence/automation-schedule.md).
Files in this skill
| File | Purpose |
|---|
skill.md | This contract |
README.md | Quick start + how it fits the ecosystem |
workflow.md | The 7-agent run, step by step |
agent-prompts.md | The prompt for each of the 7 agents |
quality-gates.md | The 6 gates a run must pass |
scoring-models.ts | The 12 formulas (pure, tested-shape, type-checked) |
types.ts | Shared types incl. ConfidenceLabel, DataStatus, Weights |
inputs.schema.json / outputs.schema.json | Run contract |
source-map.schema.json / claim-verification.schema.json | Artifact contracts |
examples.md | Worked example with placeholder→real transition |
First-run status (2026-05-29)
Run #1 output is in ../../docs/marketing-intelligence/. YouTube + live-metric deliverables are
DATA_REQUIRED because the vault contained no channel references and no GSC/Semrush exports were
present. The skill, schemas, math models, claim ledger, design docs and CEO synthesis are complete.
Source: CleanExpo/Synthex — distributed by TomeVault.