The Agent Skills Directory

Model Behavior Override Instructions: The skill uses emphatic language and directives to ensure the model prioritizes the provided documentation over its pre-trained knowledge.
Evidence: Use of markers like "CRITICAL: Always Use Correct Documentation" and "Your knowledge of workflow is outdated" (SKILL.md).
Indirect Prompt Injection Surface: The skill documentation describes patterns where user-provided messages are processed by AI agents (e.g., DurableAgent) without explicit instruction isolation.
Ingestion points: User messages are ingested in functions like myAgentWorkflow (SKILL.md) and task descriptions in agentWorkflow (SKILL.md).
Boundary markers: The documentation does not specify the use of delimiters or clear boundary markers when interpolating user data into prompts.
Capability inventory: The documented agent architecture includes tool execution, network access via fetch, and file writing (references/durable-agent-patterns.md).
Sanitization: No explicit sanitization, validation, or instruction-filtering patterns are described for user-provided data.
Command Execution for Documentation Retrieval: The skill instructs the agent to use local shell commands to navigate and search its own reference materials.
Evidence: Instructions to use glob and grep within the node_modules directory to find and read documentation (SKILL.md).
Integration with Trusted Services: The skill's external references and dependency patterns are associated with Vercel, a well-known technology service. These references are documented neutrally as part of the skill's functional purpose.

workflow