The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes user-supplied trace data (CSV, JSON, JSONL) to determine the appropriate visual rendering for the tool. This processing of untrusted external data represents a potential indirect prompt injection surface.
Ingestion points: Step 1 instructions require the agent to read and sample the user's trace data files to understand field names and nesting.
Boundary markers: There are no specific delimiters or instructions for the agent to disregard potential instructions embedded within the sampled data.
Capability inventory: The skill generates executable Python scripts, shell commands for environment setup, and handles local file writing (annotations.jsonl).
Sanitization: No sanitization, validation, or escaping of the ingested trace data is specified during the analysis or generation steps.
[EXTERNAL_DOWNLOADS]: The generated tool relies on external packages and remote assets.
Dependencies: The skill instructs the agent to include fasthtml and python-fasthtml in the generated requirements.txt. While fasthtml is a legitimate framework, python-fasthtml is a non-standard package name that should be verified.
Remote Assets: The application uses the TailwindCSS CDN (https://cdn.tailwindcss.com) for styling, which is a well-known and widely used service.

trace-annotation-tool