The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external, untrusted sources including public code repositories (via Sourcebot), third-party library documentation (via Context7), and general web research/debugging information (via Perplexity).
Ingestion points: Data enters the agent's context through the output of search and documentation retrieval tools.
Boundary markers: The instructions do not define clear delimiters or warnings to treat external content as untrusted, which increases the risk of the agent following instructions embedded in the retrieved research data.
Capability inventory: The skill has access to architectural analysis and refactoring tools (mcp__zen__*) and local codebase search.
Sanitization: There are no instructions for sanitizing or escaping the retrieved external content before it is processed or included in the final report.
[PROMPT_INJECTION]: The instructions use aggressive, authoritative language and negative constraints to override the agent's default decision-making and enforce a specific workflow and model choice.
Evidence: Phrases such as 'CRITICAL ENFORCEMENT', 'YOU MUST ALWAYS', 'NO EXCEPTIONS', 'MANDATORY', and 'YOUR TASK WILL BE INVALIDATED' are used to ensure strict adherence to the defined research steps.
Model Override: The skill explicitly mandates the use of a specific model ('anthropic/claude-opus-4.6') for certain tool calls, overriding the agent's default model selection logic.

deep-research