The Agent Skills Directory

[DATA_EXFILTRATION]: The skill includes a reporting feature (npx langwatch report) that can send session transcripts, titles, and summaries to the LangWatch team. While it requires the --user-approved flag and claims to perform local PII scrubbing (secrets, emails, IPs, etc.), this still involves sending potentially sensitive interaction data to a third-party vendor.
[DATA_EXFILTRATION]: The skill exports production traces, which include user inputs and agent outputs, to local files (traces.jsonl) for analysis. This data is sensitive and must be handled carefully within the agent's context to avoid accidental exposure.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from production traces to generate summaries and reports. This presents an attack surface where malicious content in the traces could influence the agent's findings or report output.
Ingestion points: Production traces are exported to traces.jsonl and processed by the agent.
Boundary markers: The instructions do not specify boundary markers or delimiters for the processed trace data.
Capability inventory: The skill performs file writes (HTML report) and network operations (reporting bugs via npx).
Sanitization: Local scrubbing is performed for the bug reporting tool, but no specific sanitization is mentioned for the generated HTML report.
[COMMAND_EXECUTION]: The skill uses npx to execute the langwatch package. While this is a vendor-provided resource, npx downloads and executes code from the npm registry at runtime.

agent-performance