The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs strong instructional framing ("MANDATORY", "Iron Law of Execution") in SKILL.md and replicates specific behavioral instructions (e.g., "The industry experts Principles", "Corporate Reporting: The Obsidian Loop") within asset templates and reference documents. These instructions are designed to enforce deterministic execution and specific reporting workflows rather than bypassing safety guardrails. Found in assets/campaign_report_template.md and references/funnel-optimization-framework.md.- [INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by ingesting user-provided JSON data through CLI scripts. While the scripts are limited to standard library calculations and lack dangerous capabilities, the ingestion of external data is a standard risk factor.
Ingestion points: attribution_analyzer.py, funnel_analyzer.py, and campaign_roi_calculator.py.
Boundary markers: None present in the data input schemas.
Capability inventory: Local execution of Python calculations; printing results to stdout.
Sanitization: Basic JSON parsing and type validation are present, but no explicit sanitization for downstream prompt interpolation.

campaign-analytics