The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides patterns for implementing an API that ingests and processes untrusted batches of operations. This creates a potential surface for indirect prompt injection if the inputs are not strictly sanitized.
Ingestion points: The POST /batch endpoint described in SKILL.md and references/examples.md accepts arbitrary arrays of operations.
Boundary markers: SKILL.md (Step 8) and references/implementation.md (Step 2) instruct the user to implement validation and schema enforcement, which act as boundaries.
Capability inventory: The skill utilizes allowed-tools: Bash(api:batch-*) for scaffolding and describes a system that executes operations via an executeOperation function in references/examples.md.
Sanitization: The instructions mention "input validation" and "schema enforcement" but lack specific guidance on preventing command injection or request smuggling within the batch items.
[DYNAMIC_EXECUTION]: The batch processing logic in references/examples.md implements a dynamic dispatcher. The processBatchSync and worker functions iterate over user-provided operations and dispatch them based on the path and method fields, which is a form of dynamic execution of instructions defined in user data.

processing-api-batches