The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests external, potentially untrusted data from multiple sources to inform dataset generation.
Ingestion points: Reads project files, git history, production traces via langwatch trace search, and user-provided materials (PDFs, docs).
Boundary markers: The instructions do not specify the use of delimiters or 'ignore embedded instructions' warnings when processing this external content.
Capability inventory: The agent has the ability to execute shell commands (langwatch CLI), write local files, and execute dynamically generated scripts.
Sanitization: There is no mention of sanitizing or escaping the content retrieved from production traces or codebase files before it is processed by the LLM.
[DYNAMIC_EXECUTION]: The skill instructs the agent to generate and execute local scripts for data processing tasks.
Evidence: Phase 4 ('Full Generation') explicitly directs the agent to 'Write a small Python or Node.js script' to generate CSV files rather than manually constructing strings. This is classified as a low-severity finding as it uses standard templates for data formatting.
[COMMAND_EXECUTION]: The skill frequently executes the langwatch CLI tool to interact with the vendor's platform.
Evidence: Commands include langwatch prompt list, langwatch trace search, langwatch dataset create, and langwatch dataset upload. As these operations target the skill vendor's own infrastructure (LangWatch), they are considered intended functionality but represent a significant capability tier.

datasets