Deep Ads Research

Plan and orchestrate deep research pipelines for multi-platform ads decision making across Meta (Facebook/Instagram), Google Ads, TikTok Ads, YouTube Ads, Am...

Security Scan

The skill is an instruction-only planner for ad research workflows and its declared requirements and instructions are coherent with that purpose.

Purpose & Capability

Name/description (deep ads research and orchestration across ad platforms) match the SKILL.md content. The skill is a planning/orchestration tool and does not declare any binaries, installs, or credentials that would be unexpected for that role.

Instruction Scope

SKILL.md contains step-by-step guidance for decomposing research questions, defining sources, synthesizing evidence, and producing decisions. It does not instruct the agent to read system files, use credentials, hit external endpoints, or perform platform-side actions directly. Mentions of 'platform_data' or 'internal_tests' are sources to collect evidence but are not implemented as automated data access steps in the instructions.

Install Mechanism

Instruction-only skill with no install spec and no code files. No artifacts are downloaded or written to disk, minimizing install-related risk.

Credentials

No required environment variables, credentials, or config paths are declared. The SKILL.md does not ask for secrets or unrelated credentials. If the agent later requests platform credentials from a user, that would be external to the skill's declared requirements and should be reviewed before sharing.

Persistence & Privilege

The skill is not marked always:true and is user-invocable. disable-model-invocation is false (the default), so the agent may invoke the skill autonomously when eligible — this is normal for skills. Because this skill does not request credentials or write installs, autonomous invocation has low additional risk, but you may still want to control when it runs.

Assessment

This skill is a coherent planner for ad research and does not itself request credentials or install code. Before providing any platform data or credentials to the agent (e.g., API keys, campaign exports, or internal test data), confirm you trust the agent session and limit scope to the minimum needed. If you expect the skill to perform live actions (creating campaigns, changing bids), require explicit authorization steps and avoid storing long-lived credentials. If you want to further reduce risk, disallow autonomous invocation or require manual approval before the agent uses this skill.

Purpose

Core mission:

• research workflow orchestration, source plan, synthesis output

This skill is specialized for advertising workflows and should output actionable plans rather than generic advice.

When To Trigger

Use this skill when the user asks for:

• ad execution guidance tied to business outcomes
• growth decisions involving revenue, roas, cpa, or budget efficiency
• platform-level actions for: Meta (Facebook/Instagram), Google Ads, TikTok Ads, YouTube Ads, Amazon Ads, DSP/programmatic
• this specific capability: research workflow orchestration, source plan, synthesis output

High-signal keywords:

• ads, advertising, campaign, growth, revenue, profit
• roas, cpa, roi, budget, bidding, traffic, conversion, funnel
• meta, googleads, tiktokads, youtubeads, amazonads, shopifyads, dsp

Input Contract

Required:

• research_question
• hypothesis_set
• decision_deadline

Optional:

• source_preferences
• confidence_target
• excluded_assumptions
• output_depth

Output Contract

1. Research Plan
2. Evidence Table
3. Hypothesis Evaluation
4. Strategic Conclusion
5. Actionable Next Experiments

Workflow

1. Decompose research question into testable hypotheses.
2. Define source and evidence collection plan.
3. Evaluate evidence strength and conflicts.
4. Synthesize implications for ad strategy.
5. Output decisions and follow-up experiments.

Decision Rules

• If evidence quality is weak, state limitation and avoid hard claims.
• If hypotheses conflict, rank by evidence strength and recency.
• If decision deadline is near, provide best-effort recommendation with risk notes.

Platform Notes

Primary scope:

• Meta (Facebook/Instagram), Google Ads, TikTok Ads, YouTube Ads, Amazon Ads, DSP/programmatic

Platform behavior guidance:

• Keep recommendations channel-aware; do not collapse all channels into one generic plan.
• For Meta and TikTok Ads, prioritize creative testing cadence.
• For Google Ads and Amazon Ads, prioritize demand-capture and query/listing intent.
• For DSP/programmatic, prioritize audience control and frequency governance.

Constraints And Guardrails

• Never fabricate metrics or policy outcomes.
• Separate observed facts from assumptions.
• Use measurable language for each proposed action.
• Include at least one rollback or stop-loss condition when spend risk exists.

Failure Handling And Escalation

• If critical inputs are missing, ask for only the minimum required fields.
• If platform constraints conflict, show trade-offs and a safe default.
• If confidence is low, mark it explicitly and provide a validation checklist.
• If high-risk issues appear (policy, billing, tracking breakage), escalate with a structured handoff payload.

Code Examples

Research Plan YAML

hypothesis: creator-led videos improve roas in week 1
sources: [platform_data, competitor_examples, internal_tests]
confidence_target: medium_high

Evidence Row

source: campaign_2026_q1
finding: cpa_down_18pct
confidence: medium

Examples

Example 1: Deep competitor study

Input:

• Need three-month competitor creative and offer shifts
• Channels: Meta + TikTok Ads

Output focus:

• evidence table
• pattern summary
• strategic implications

Example 2: Hypothesis stress test

Input:

• Team believes broad targeting always wins
• Evidence is mixed

Output focus:

• hypothesis decomposition
• confidence-ranked conclusions
• follow-up experiments

Example 3: Board-level strategic brief

Input:

• Need recommendation for next quarter channel direction
• Budget increases available

Output focus:

• scenario options
• risk-weighted recommendation
• decision-ready summary

Quality Checklist

• Required sections are complete and non-empty
• Trigger keywords include at least 3 registry terms
• Input and output contracts are operationally testable
• Workflow and decision rules are capability-specific
• Platform references are explicit and concrete
• At least 3 practical examples are included