The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes performance logs (perf.jsonl) which can contain arbitrary strings from UI attributes (e.g., item IDs, repo names).
Ingestion points: The skill instructs the agent to read and analyze perf.jsonl via several jq command patterns in Section 4.
Boundary markers: Absent. There are no instructions for the agent to treat the contents of the performance logs as untrusted or to ignore potential instructions embedded within those logs.
Capability inventory: The skill has shell execution capabilities including jq, grep, wc, and the mise task runner.
Sanitization: Absent. The skill does not provide methods to sanitize or escape the contents of the JSONL file before the agent processes the results.
[COMMAND_EXECUTION]: The skill utilizes multiple shell commands to interact with the local development environment and analyze data.
Evidence: Uses grep and cut on launch.env to identify the data directory and verify performance recording settings.
Evidence: Uses mise run commands to prepare and serve the development environment.
Evidence: Uses jq with complex filters to aggregate, sort, and display performance metrics based on duration and attributes.

desktop-ui-perf