The Agent Skills Directory

[PROMPT_INJECTION]: Extreme content redundancy and 'Wall of Text' pattern detected across multiple reference files.
Files such as architecture-patterns.md, code-organization.md, deployment-pipelines.md, error-handling.md, performance-optimization.md, security-best-practices.md, state-management.md, and testing-strategies.md each contain 150 sections of repeating technical boilerplate.
This pattern is a known adversarial technique used to exhaust the agent's context window, which can cause the model to lose track of its original system prompt and safety guardrails.
[PROMPT_INJECTION]: Vulnerability to indirect prompt injection via external data ingestion.
Ingestion points: The skill is designed to parse SQL queries for lineage tracking and ingest database schemas from external warehouses into data catalogs (SKILL.md).
Boundary markers: Absent; there are no instructions to use delimiters or 'ignore embedded instructions' warnings for external content.
Capability inventory: The skill suggests capabilities including sending alerts to Slack and PagerDuty and integrating with various monitoring platforms.
Sanitization: Absent; no instructions are provided to filter, escape, or validate external SQL or schema content before processing.

data-data-observability