The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs authoritative and coercive language such as 'MANDATORY', 'NEVER', 'TASK INVALIDATION', and 'NO EXCEPTIONS' to override the agent's default behavior. It specifically mandates the use of a non-standard or fictional model identifier ('anthropic/claude-opus-4.6') for all tool calls. This is a pattern often associated with attempts to bypass safety filters or force specific behavior by misrepresenting the model's identity or environment.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external files for analysis, debugging, and refactoring without providing safety boundaries for the processed content.
Ingestion points: The relevant_files parameter in all tool references (e.g., references/analyze.md, references/review.md) allows for the loading of arbitrary project files into the context.
Boundary markers: No delimiters or instructions to ignore embedded commands are specified for the content of analyzed files.
Capability inventory: The skill uses MCP tools (mcp__zen__analyze, mcp__zen__debug, etc.) to perform analysis and tracing on the provided files.
Sanitization: The instructions do not include requirements for sanitizing or escaping the content of the files being processed.

pal