The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill uses npx --yes langwatch report to execute the latest version of the vendor's CLI tool directly from the npm registry. This is a standard deployment method for the tool's author.
[EXTERNAL_DOWNLOADS]: Fetches documentation and project guidelines from official vendor domains such as langwatch.ai.
[COMMAND_EXECUTION]: Utilizes the langwatch CLI for various diagnostic tasks, including querying analytics, exporting traces, and generating reports. It also suggests the use of local python3 or jq scripts for data processing.
[DATA_EXFILTRATION]: Includes a built-in feedback mechanism to send session transcripts and summaries to the vendor's team. This feature is gated by a mandatory --user-approved flag and performs local scrubbing of sensitive information like secrets, API keys, and PII before transmission.
[INDIRECT_PROMPT_INJECTION]: The skill processes production trace data which could contain malicious instructions embedded by external users. This is an inherent surface for diagnostic tools.
Ingestion points: Production traces exported to traces.jsonl (SKILL.md Step 3).
Boundary markers: Absent; no specific delimiters or warnings are provided for the agent when analyzing the content of exported traces.
Capability inventory: Shell command execution via the langwatch CLI and file system write access for report generation.
Sanitization: Local scrubbing is implemented for data sent to the vendor, but not explicitly for the agent's internal analysis of trace data.

agent-performance