TubeSentry Agent Skill (youtube-placement-audit)

TubeSentry includes an installable Agent Skill compliant with modern coding assistant formats (Antigravity, Claude Code, Cursor, Codex).

The skill gives autonomous AI agents step-by-step decision rules, tool calling recipes, data contracts, and operational best practices for auditing YouTube and Performance Max placement reports.

---

Installing the Skill

1. Via skills.sh / npx (Universal Agent Installer)

Install directly to Claude Code, Cursor, Copilot, or Antigravity via the open skills ecosystem:

npx skills add flovv/tubesentry

2. Via TubeSentry CLI

Run the CLI command in any workspace root:

tubesentry skill install

Or without installing Python packages locally:

uvx --from tubesentry tubesentry skill install

The installer automatically detects the workspace environment:

1. ./.agents/skills/ (Antigravity / Hermes standard)
2. ./.claude/skills/ (Claude Code standard)
3. Custom location via --target-dir <path>

---

Inspecting the Skill Content

You can print the full skill guide directly to the terminal:

tubesentry skill show

Or query the local filesystem path of the bundled markdown file:

tubesentry skill path

---

MCP Server Integration

Agents connected via the FastMCP server can also query the skill instructions dynamically via the get_audit_skill_guide tool:

from tubesentry.mcp_server import get_audit_skill_guide

guide = get_audit_skill_guide()

---

What the Skill Teaches the Agent

• Division of Responsibilities: Never write ad-hoc regex parsing or manual CSV loops; always use audit_placements() and native quantizers.
• Metric Hierarchy: Prioritize impression-weighted quality scores over raw channel counts (e.g. 10 toddler channels burning 45% of impressions).
• Creative Alignment: How to contrast multimodal Gemini ICP targeting intent with actual algorithmic delivery.
• Google Ads Constraints: Applying the 65,000-channel exclusion limit correctly via Google Ads Editor shields.